AI Tutorial: Using Text Embeddings to Label Synthetic Doctor’s Notes Generated with ChatGPT

I’ve been playing around with OpenAI and ChatGPT in my research, and I thought I’d put together a short tutorial that demonstrates using ChatGPT API to generate synthetic doctor’s notes, and then using OpenAI’s text embedding models to label the notes according to whether they involve a chronic or acute condition. And yes, I’m fully aware that what I write here will probably be out of date in about 3 hours.
data science
generative AI
ChatGPT
OpenAI
machine learning
Author

Rebecca Barter

Published

April 11, 2024

An image created by dall-e-3 using the name of this blog post as the prompt. The image contains a brain in the center connected using circuitry to a code console, with various viruses and other things floating around, and a doctor conducting a medical exam of what seems to be another doctor

Warning: If you’re reading this post more than 3 hours after it was posted, some (maybe all) of the information it contains will probably be out of date.

Hello friends, welcome to the new rapidly evolving world of AI. Even the wacky image that you see at the top of this blog was created using AI (specifically, using dall-e-3), just for this blog post!

Like me, I’m sure you’re dwelling on the question of whether AI is going to take all of our jobs. I think, for now, I’ve settled on the opinion that for those of us who adapt and learn to work with AI, our jobs are pretty safe. However, there is no question that AI is going to fundamentally change our jobs, hopefully for the better. In my day-to-day I have already found that GitHub Copilot has boosted my coding efficiency, chatGPT has boosted my efficiency for mundane writing tasks, and the various other ways that AI has snuck into my workday are probably helping me out in other ways too.

And, as you’ll see in this blog post, using the tools provided by OpenAI, NLP tasks like generating synthetic text, and labeling mass amounts of text data are now fairly straightforward too.

In this blog post, I will walk through an example of generating a small collection of synthetic doctor’s notes using chatGPT and then using text OpenAI’s embedding models to automatically label these doctor’s notes in terms of whether they correspond to a patient experiencing a chronic or an acute event. Fortunately, the documentation provided on the OpenAI website is very good, and that will certainly be the best point of reference.

Note that, for now, if you want to use these tools effectively, you’ll want to be fairly proficient in Python (if you’re an R user who is interested in learning Python, check out my “An introduction to Python for R Users” blog post).

Getting set up

As usual, the first thing you’ll want to do is to import the libraries you’ll need.

import pandas as pd
from scipy.spatial.distance import cosine
from openai import OpenAI

pd.set_option('display.max_colwidth', None)

Then you’ll need to set up your OpenAI API key. An API key is literally a jumble of letters that OpenAI uses to identify you so that they can track your usage. After you create an OpenAI account, see the OpenAI API keys website for information on creating an API key.

Once you have created an API key, and you have saved it somewhere on your computer, say in a file called “api_key.txt”, you will need to use it to connect to the OpenAI client.

To do this, either create a local variable called your_key or read in your key from your text file using code like the following:

with open('api_key.txt', 'r') as file:
  your_key = file.read().strip()

Then you can set up your OpenAI client which is how you will connect to the OpenAI API:

# Define a local variable called `your_key` that contains your OpenAI key.
# Try really hard to avoid writing your key in your notebook
# And when you do, try even harder to avoid uploading it to GitHub
client = OpenAI(api_key=your_key)

Note that it is really, really important that you do not write your API key in plain text in your notebook file, especially if you are going to upload your notebook publicly to GitHub. This would allow someone else to pretend to be you and use your OpenAI account.

Instead, you will want to only ever define your key in a local variable (that is not saved in your code) or load it in from a file that should not be uploaded to GitHub!

Once you’re set up, let’s generate some fake doctor’s notes to analyze!

Creating synthetic doctor’s notes with chatGPT

The first thing I did was use chatGPT-4 to create a collection of synthetic doctor’s notes.

Note that at the time of writing, you have to pay to access chatGPT-4 through the API. If you don’t want to pay OpenAI, you can replace model="gpt-4" with model="gpt-3.5-turbo", which is currently free (but is also less good).

The following code uses chatGPT-4 to generate 50 doctor’s notes using the prompt:

“Provide a collection of 50 doctors notes that resemble the kind that you would enter into an EHR for your patients on a day-to-day basis. Each note should have around 3 sentences. Each note should appear on a new line. Do not add any numbers or superfluous text.”

completion = client.chat.completions.create(
  model="gpt-4",
  messages=[
    {"role": "system", 
     "content": "You are a doctor working in a large hospital."},
    {"role": "user",
     "content": """Provide a collection of 50 doctors notes that 
                   resemble the kind that you would enter into an 
                   EHR for your patients on a day-to-day basis. 
                   Each note should have around 3 sentences. 
                   Each note should appear on a new line.
                   Do not add any numbers or superfluous text."""}
  ]
)

The text created by ChatGPT can be extracted using completion.choices[0].message.content. But since this is just one big string value, I want to use the .split method to separate each note into a list element.

# extract the notes and place them in a list, where each list element is detected by the presence of a line break ('\n')
notes_list = completion.choices[0].message.content.split('\n')

Since the output provided tended to add additional blank lines between the notes (even when I asked it not to), I used the following code to remove any entries in my notes_list that contain blank strings:

# remove any empty notes
# may or may not be required for you
notes_list = [note for note in notes_list if note != '']
# place the notes in a DataFrame
notes_df = pd.DataFrame({'notes': notes_list})

Let’s take a look at the notes it created:

# look at the first 5
notes_df.head(5)
notes
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms.
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause.
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs.
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise.
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated.

However, since chatGPT returns a different collection of notes every time, you can load the notes I created above using the following code (if you prefer to follow along with my doctor’s notes, that is):

notes_df = pd.read_csv("https://raw.githubusercontent.com/rlbarter/personal-website-quarto/main/blog/data/doctors_notes.csv", index_col=0)

If you just loaded the notes using the URL above, take a look at the first 5 notes.

# look at the first 5
notes_df.head(5)
notes
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms.
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause.
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs.
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise.
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated.

Computing text embeddings

My goal in the rest of this post will be to identify whether each note corresponds to a patient with a chronic condition (which I define as a condition that they have been experiencing for more than one month) or an acute condition (e.g., a new condition).

To do this, I will use something called text embeddings. The idea is that I will use OpenAI’s pre-trained text embedding models to embed each note into a numeric 1,536-dimensional that somehow approximately respects semantic distance.

The equivalent of this in a two-dimensional world would be taking each individual doctor’s note and placing it on a scatterplot so that notes that we might intuitively consider to be more “similar” to one another are closer together in the plot, and notes that we might consider to be more “different” from one another are further apart. Unfortunately, it is unlikely that we would be able to come up with any two quantifiable values that we could use to define the two axes of our scatterplot, such that when we place each note in the scatterplot we achieve our desired property that “similar” notes are closer together. But it turns out that once your space has, say, 1,536 dimensions, this task somehow becomes possible (although what each dimension/axis represents is not necessarily going to be meaningful to us).

Fortunately, OpenAI has already trained some general text embedding models that we can use to embed each of the doctor’s notes in such a 1,536-dimensional space. In this post, I will be using the “text-embedding-3-small” OpenAI model to compute the text embeddings.

After computing the embeddings, our task becomes determining how close each doctor’s note is to the embedding of the following text: “Patient presenting with ongoing chronic condition defined as ongoing for more than one month”, which I will call the target. The closer a note’s embedding is to the target’s embedding, the more likely the note corresponds to a patient who is presenting with a chronic condition!

I hear you. That sounds crazy, right? But it works… eerily well! Let me show you.

First, I need to compute the embeddings for each of the chatGPT-generated doctor’s notes. I can do that using this custom function get_embedding() function that I literally just copy-and-pasted from the OpenAI documentation:

def get_embedding(text, model="text-embedding-3-small"):
   text = text.replace("\n", " ")
   return client.embeddings.create(input = [text], model=model).data[0].embedding

This function basically takes a string text, and uses the “text-embedding-3-small” model to create a list of 1,536 numbers corresponding to the 1,536-dimensional embedding of the text entry.

For example, the following code creates an embedding of the text “hello, how are you”:

get_embedding('hello, how are you')
[0.020681990310549736,
 -0.03974681720137596,
 -0.000452458014478907,
 0.029285797849297523,
 -0.013973292894661427,
 -0.059632860124111176,
 0.0007248803740367293,
 0.02625361829996109,
 0.0012578805908560753,
 -0.039999499917030334,
 0.004143978469073772,
 -0.032242171466350555,
 -0.011370671913027763,
 -0.04505313187837601,
 -0.005956968758255243,
 0.06473702937364578,
 -0.02131369337439537,
 0.05190080404281616,
 -0.0056916531175374985,
 0.02920999377965927,
 0.058015696704387665,
 0.01499665342271328,
 0.009684022516012192,
 0.04065646976232529,
 0.0349964015185833,
 0.025268161669373512,
 0.023827875033020973,
 0.01146542839705944,
 0.02449748106300831,
 -0.017624542117118835,
 0.05356850102543831,
 -0.03800331428647041,
 0.012665665708482265,
 0.013291052542626858,
 0.002278872299939394,
 0.013834318146109581,
 -0.0029911184683442116,
 0.009267098270356655,
 -0.001953544793650508,
 -0.060896266251802444,
 -0.006733964662998915,
 -0.01929224096238613,
 0.0412629060447216,
 0.016411669552326202,
 0.01106745470315218,
 0.0036165055353194475,
 0.003145886119455099,
 -0.018572097644209862,
 0.04836326092481613,
 0.020088188350200653,
 0.02371416985988617,
 -0.01297520101070404,
 0.014529192820191383,
 0.13715557754039764,
 0.04500259459018707,
 -0.04282953217625618,
 0.036563027650117874,
 0.00012782136036548764,
 0.005268411710858345,
 0.03388460353016853,
 0.002237811451777816,
 -0.04341069981455803,
 0.01293729804456234,
 0.027264345437288284,
 -0.0006838195840828121,
 -0.030422866344451904,
 -0.013758513145148754,
 0.012728836387395859,
 -0.016828594729304314,
 3.400344212423079e-05,
 0.02913418971002102,
 0.04070700705051422,
 -0.008610125631093979,
 -0.018572097644209862,
 -0.03648722544312477,
 0.016007380560040474,
 -0.0004595646751113236,
 -0.032823339104652405,
 0.01567889377474785,
 0.02673371322453022,
 -0.0628671869635582,
 -0.023524656891822815,
 -0.03421308845281601,
 -0.006399161648005247,
 -0.032444316893815994,
 0.010827407240867615,
 -0.014794507995247841,
 0.01747293397784233,
 0.005975920241326094,
 0.004292428959161043,
 -0.02663264237344265,
 0.012634080834686756,
 -0.02045457623898983,
 0.005565312225371599,
 0.022488662973046303,
 -0.04763048142194748,
 4.67115496576298e-05,
 0.02693585865199566,
 0.036360882222652435,
 -0.00024103456235025078,
 0.05119329318404198,
 -0.02443431131541729,
 -0.0230319295078516,
 -0.002187275094911456,
 0.05336635559797287,
 0.03901404142379761,
 0.021717984229326248,
 -0.020290333777666092,
 -0.014124901965260506,
 0.011219063773751259,
 -0.10142640024423599,
 -0.002395737450569868,
 -0.012570910155773163,
 0.006550770718604326,
 0.04090915247797966,
 0.03982262313365936,
 0.05958232283592224,
 -0.04581117630004883,
 0.025470305234193802,
 -0.02701166458427906,
 0.007056133821606636,
 -0.015982111915946007,
 0.03939306363463402,
 -0.04042905569076538,
 -0.061755385249853134,
 -0.010941113345324993,
 -0.014491289854049683,
 0.02185695990920067,
 -0.03727053850889206,
 -0.006525502540171146,
 -0.008875441737473011,
 0.023082464933395386,
 0.009854583069682121,
 0.0023704692721366882,
 -0.028376145288348198,
 0.013152077794075012,
 -0.020745160058140755,
 0.03964574262499809,
 -0.045659568160772324,
 0.014327047392725945,
 0.061654310673475266,
 -0.04992988705635071,
 -0.013632172718644142,
 0.01441548578441143,
 0.005697970278561115,
 -0.06215967610478401,
 -0.0039228820241987705,
 0.04992988705635071,
 0.010953747667372227,
 -0.014251242391765118,
 0.03524908423423767,
 0.012861493974924088,
 -0.0747937560081482,
 0.023309879004955292,
 0.002479438204318285,
 0.008066860027611256,
 0.030448133125901222,
 -0.021086279302835464,
 -0.023486755788326263,
 0.03287387639284134,
 0.013000468723475933,
 0.0094060730189085,
 -0.03125671669840813,
 -0.04914657399058342,
 -0.029285797849297523,
 0.00018635268497746438,
 -0.021490570157766342,
 0.04257684946060181,
 0.0014047517906874418,
 0.017422396689653397,
 -0.03153466433286667,
 0.03924145549535751,
 -0.04664502665400505,
 -0.02923526242375374,
 0.0100061921402812,
 -0.02749175950884819,
 0.028805702924728394,
 -0.01360690500587225,
 -0.05020783469080925,
 0.003398567670956254,
 0.023865777999162674,
 0.02221071347594261,
 0.019608093425631523,
 -0.015249335207045078,
 -0.013291052542626858,
 0.016310598701238632,
 -0.025167088955640793,
 0.010656846687197685,
 -0.028199266642332077,
 0.017460299655795097,
 0.010770553722977638,
 0.0555899553000927,
 0.004349282011389732,
 0.018799511715769768,
 0.025697719305753708,
 -0.04399186745285988,
 0.07878612726926804,
 -0.0081489821895957,
 -0.04550795629620552,
 0.023385683074593544,
 -0.008824905380606651,
 -0.008212151937186718,
 0.00013364487676881254,
 -0.02683478593826294,
 0.0020340869668871164,
 0.007744691334664822,
 -0.02515445463359356,
 0.06332200765609741,
 0.008654344826936722,
 -0.0013494776794686913,
 0.04816111549735069,
 -0.020568283274769783,
 0.003951308783143759,
 0.04667029157280922,
 -0.009046001359820366,
 -0.03171154111623764,
 0.04404240474104881,
 0.03380879759788513,
 0.00862907711416483,
 0.004924132954329252,
 -0.06377683579921722,
 0.04295587167143822,
 0.08070650696754456,
 -0.02221071347594261,
 0.026278886944055557,
 0.012071863748133183,
 0.027087468653917313,
 0.005540044046938419,
 0.013316321186721325,
 -0.003521749982610345,
 0.007605716586112976,
 0.028780434280633926,
 0.025621915236115456,
 -0.021440034732222557,
 -0.03997423127293587,
 0.011484378948807716,
 0.015779966488480568,
 -0.04282953217625618,
 0.019140630960464478,
 -0.00490834005177021,
 -0.021819056943058968,
 -0.026657909154891968,
 -0.02341095171868801,
 0.017510835081338882,
 0.03393514081835747,
 -0.005634800065308809,
 0.021149450913071632,
 -0.0350722074508667,
 -0.00835744384676218,
 0.011901304125785828,
 -0.003910247702151537,
 0.01939331367611885,
 0.014781873673200607,
 -0.041338711977005005,
 -0.003625981044024229,
 -0.002013556659221649,
 -0.008231103420257568,
 0.004658817313611507,
 0.00835744384676218,
 -0.02357519418001175,
 -0.028881506994366646,
 -0.037523217499256134,
 -0.008104762993752956,
 -0.03054920583963394,
 -0.021149450913071632,
 -0.045836444944143295,
 -0.015552553348243237,
 0.01571679674088955,
 0.012002376839518547,
 0.02059355191886425,
 0.04214729368686676,
 -0.0374726839363575,
 0.03459211066365242,
 0.01058735977858305,
 -0.022981392219662666,
 0.02759283222258091,
 0.02728961408138275,
 0.008679613471031189,
 0.03075135126709938,
 -0.008603808470070362,
 -0.0325959287583828,
 -0.009791412390768528,
 0.003357506822794676,
 -0.02835087664425373,
 -0.043056946247816086,
 0.030877692624926567,
 -0.024901771917939186,
 -0.02213490940630436,
 0.035678643733263016,
 0.019191168248653412,
 0.01446602214127779,
 -0.00023274344857782125,
 -0.002964271232485771,
 -0.028805702924728394,
 0.0007979211513884366,
 0.01111799106001854,
 0.036739904433488846,
 -0.049980420619249344,
 0.01627269573509693,
 -0.012949932366609573,
 0.052456703037023544,
 0.004175563808530569,
 0.013240516185760498,
 -0.01747293397784233,
 -0.0157546978443861,
 0.04368865117430687,
 0.021402131766080856,
 -0.0054516056552529335,
 0.0010115160839632154,
 0.0004481150535866618,
 -0.03982262313365936,
 0.03297495096921921,
 0.05766194313764572,
 -0.02741595357656479,
 -0.014604996889829636,
 -0.010833724401891232,
 0.02855302207171917,
 0.0026926384307444096,
 0.004750414285808802,
 -0.034263625741004944,
 0.020201895385980606,
 -0.0313577875494957,
 -0.016866497695446014,
 -0.018812146037817,
 -0.032444316893815994,
 -0.014289145357906818,
 0.024825967848300934,
 -0.003780748462304473,
 -0.04176827147603035,
 -0.021490570157766342,
 -0.0008843856048770249,
 -0.06908315420150757,
 0.006279137916862965,
 -0.005394752137362957,
 0.04414347559213638,
 0.008717515505850315,
 -0.03236851468682289,
 0.033303435891866684,
 0.002240970032289624,
 0.014920849353075027,
 -0.015186164528131485,
 -0.0399489626288414,
 -0.019936578348279,
 -0.051218561828136444,
 -0.003907089587301016,
 -0.019090095534920692,
 0.01955755613744259,
 0.002414688700810075,
 -0.031281981617212296,
 -0.021920129656791687,
 0.018989022821187973,
 0.01154754962772131,
 -0.0017545579466968775,
 -0.027643367648124695,
 -0.012659348547458649,
 -0.0206314530223608,
 0.07332820445299149,
 -0.0028174000326544046,
 -0.013253150507807732,
 0.025280794128775597,
 0.0007900248165242374,
 -0.008420614525675774,
 2.5046078008017503e-05,
 -0.024067923426628113,
 0.04245050996541977,
 0.022804515436291695,
 -0.06023929640650749,
 -0.008572223596274853,
 0.018445758149027824,
 0.049197107553482056,
 0.025204990059137344,
 0.0005176024860702455,
 0.0021714826580137014,
 0.03476898744702339,
 -0.06498970836400986,
 0.02979116141796112,
 0.037220001220703125,
 0.004434562288224697,
 0.0020909402519464493,
 0.011724426411092281,
 0.0931384414434433,
 0.03727053850889206,
 -0.013948025181889534,
 -0.025722987949848175,
 0.027062200009822845,
 0.00029196569812484086,
 -0.009791412390768528,
 -0.006696062628179789,
 -0.03974681720137596,
 0.033758264034986496,
 0.08611389249563217,
 -0.012867811135947704,
 0.015514650382101536,
 -0.037801168859004974,
 0.0338340662419796,
 0.007668886799365282,
 0.028022389858961105,
 0.0259756688028574,
 -0.04942452162504196,
 -0.04308221489191055,
 0.03171154111623764,
 -0.055135127156972885,
 0.009551364928483963,
 -0.004949401132762432,
 0.014807142317295074,
 0.035880789160728455,
 -0.03792750835418701,
 -0.011105356737971306,
 0.04760521650314331,
 0.012166619300842285,
 -0.0026942177210003138,
 -0.05745979771018028,
 0.03643668815493584,
 0.060694120824337006,
 -0.010220970958471298,
 -0.018281513825058937,
 -0.027820244431495667,
 -0.02971535734832287,
 -0.0034522623755037785,
 -0.0050504738464951515,
 0.025167088955640793,
 0.012318228371441364,
 0.026607373729348183,
 0.02683478593826294,
 -0.031004033982753754,
 0.019026925787329674,
 0.025760889053344727,
 -0.04502786323428154,
 0.03600712865591049,
 -0.06888100504875183,
 -0.03355611860752106,
 0.0016124245012179017,
 0.009835631586611271,
 0.014327047392725945,
 0.009785095229744911,
 -0.017864590510725975,
 0.02855302207171917,
 0.015640990808606148,
 0.013569002039730549,
 0.003704944159835577,
 0.029563747346401215,
 -0.027087468653917313,
 -0.037801168859004974,
 -0.048060040920972824,
 -0.009349219501018524,
 -0.01725815422832966,
 0.004706195089966059,
 0.020151358097791672,
 -0.01653801091015339,
 -0.012810957618057728,
 -0.039797354489564896,
 0.03161047026515007,
 0.013303686864674091,
 0.018243612721562386,
 0.00651918537914753,
 0.01388485450297594,
 -0.027138004079461098,
 -0.006377052050083876,
 0.017662445083260536,
 -0.005874847527593374,
 -0.0029579540714621544,
 -0.00447878148406744,
 -0.034920599311590195,
 0.06140163168311119,
 0.00862907711416483,
 -0.008003690280020237,
 -0.012539324350655079,
 -0.004687243606895208,
 -0.005878006108105183,
 -0.011730743572115898,
 0.003411201760172844,
 0.003979735542088747,
 0.044320352375507355,
 0.013960658572614193,
 -0.03719473257660866,
 -0.03140832483768463,
 -0.027466490864753723,
 0.020543014630675316,
 0.013240516185760498,
 -0.02327197603881359,
 -0.013695343397557735,
 0.02025243081152439,
 0.0024004753213375807,
 0.04816111549735069,
 -0.01360690500587225,
 0.019519655033946037,
 -0.002253604121506214,
 0.014832410030066967,
 -0.06049197539687157,
 0.03178734704852104,
 0.018129905685782433,
 0.045634299516677856,
 0.04022691026329994,
 -0.013695343397557735,
 -0.03676517307758331,
 0.05013203248381615,
 -0.03724526986479759,
 -0.00802895799279213,
 -0.007321449462324381,
 -0.019759701564908028,
 -0.00988616794347763,
 -0.024585921317338943,
 -0.000546818773727864,
 0.002975326031446457,
 0.07534965872764587,
 0.019987115636467934,
 -0.007908934727311134,
 -0.02797185443341732,
 0.004289270378649235,
 -0.04237470403313637,
 0.005915908142924309,
 0.006209650542587042,
 -0.02289295382797718,
 -0.016222158446907997,
 -0.007005597464740276,
 0.02595040202140808,
 0.01771298050880432,
 0.03219163790345192,
 -0.010239922441542149,
 -0.050814270973205566,
 0.025647183880209923,
 0.010378897190093994,
 0.018824780359864235,
 -0.013392125256359577,
 -0.02563454955816269,
 0.006449698004871607,
 -0.05948125198483467,
 -0.027921317145228386,
 -0.01785195618867874,
 0.018167806789278984,
 0.014693435281515121,
 -0.03727053850889206,
 -0.02776970900595188,
 0.012596177868545055,
 -0.00855958927422762,
 -0.013569002039730549,
 0.055893171578645706,
 -0.02739068679511547,
 0.006822403520345688,
 -0.06468649208545685,
 0.014314413070678711,
 0.04141451418399811,
 -0.003040075534954667,
 -0.014377583749592304,
 -0.007902617566287518,
 0.01255827583372593,
 0.01733395829796791,
 -0.013720611110329628,
 0.014554460532963276,
 0.014440753497183323,
 0.018951119855046272,
 -0.0007438315078616142,
 -0.015097726136446,
 0.004245051182806492,
 0.004570378456264734,
 0.006569721736013889,
 -0.02741595357656479,
 -0.03168627247214317,
 0.023688901215791702,
 0.03547649830579758,
 -0.03229270875453949,
 -0.014743971638381481,
 -0.005846420768648386,
 0.0059853955172002316,
 0.006525502540171146,
 0.04712511971592903,
 -0.0026926384307444096,
 -0.0008496419177390635,
 -0.004990461748093367,
 -0.011692841537296772,
 -0.04543215408921242,
 0.0009396597160957754,
 -0.013177345506846905,
 -0.08060543239116669,
 0.03340451046824455,
 0.007706788834184408,
 0.00399868655949831,
 0.021187352016568184,
 -0.02203383669257164,
 0.016121085733175278,
 -0.012463520281016827,
 -0.03840760514140129,
 -0.022147543728351593,
 0.0014797666808590293,
 -0.03643668815493584,
 0.030448133125901222,
 0.0022630796302109957,
 0.014655533246695995,
 0.028856240212917328,
 0.05250723659992218,
 0.0012602495262399316,
 -0.0028237169608473778,
 0.029058385640382767,
 -0.02749175950884819,
 0.015640990808606148,
 -0.04328436031937599,
 -0.032722268253564835,
 -0.02883097156882286,
 -0.006446539424359798,
 0.011162210255861282,
 -0.023082464933395386,
 -0.0059032742865383625,
 -0.0058053601533174515,
 -0.0521029494702816,
 0.02055564895272255,
 -0.003995527978986502,
 -0.019759701564908028,
 0.009317634627223015,
 -0.024585921317338943,
 -0.049879349768161774,
 -0.021048378199338913,
 -0.003600712865591049,
 0.01647484116256237,
 -0.04368865117430687,
 0.0026452606543898582,
 0.019671263173222542,
 -0.019683897495269775,
 0.013227881863713264,
 -0.0014758185716345906,
 -0.0006309144082479179,
 -0.042399972677230835,
 0.009159708395600319,
 0.016424303874373436,
 0.021528473123908043,
 0.03423835709691048,
 0.01139594055712223,
 -0.018685804679989815,
 -0.019279606640338898,
 -0.03843287378549576,
 0.016904398798942566,
 0.01715708151459694,
 0.01257722731679678,
 0.00628861365839839,
 -0.01911536417901516,
 -0.008332176133990288,
 -0.012387716211378574,
 0.0033827750012278557,
 -0.02645576372742653,
 0.04495205730199814,
 -0.03269699960947037,
 0.024863870814442635,
 0.0015310925664380193,
 0.019443849101662636,
 -0.023448852822184563,
 0.029816430062055588,
 0.006396003067493439,
 -0.013922756537795067,
 0.0100061921402812,
 -0.016386402770876884,
 -0.01237508188933134,
 0.031105106696486473,
 -0.026379959657788277,
 -0.020025016739964485,
 -0.028502484783530235,
 0.004845169838517904,
 0.021098913624882698,
 -0.042197827249765396,
 0.019848139956593513,
 -0.0049367668107151985,
 0.0067402818240225315,
 0.014743971638381481,
 0.03734634071588516,
 0.013126809149980545,
 0.007195108570158482,
 -0.005631641484797001,
 0.019709166139364243,
 0.028325608000159264,
 -0.012962566688656807,
 -0.006165431346744299,
 -0.01204659603536129,
 -0.013758513145148754,
 -0.004668292589485645,
 0.05265884846448898,
 -0.037017855793237686,
 0.036259811371564865,
 -0.014301778748631477,
 -0.014390217140316963,
 -0.0026373642031103373,
 -0.0060296147130429745,
 -0.023549925535917282,
 -0.03661356493830681,
 0.018205709755420685,
 -0.034162554889917374,
 -0.002649998292326927,
 0.0009459767607040703,
 0.038938235491514206,
 0.06498970836400986,
 -0.016727522015571594,
 -0.02779497765004635,
 0.0223370548337698,
 -0.018951119855046272,
 -0.0023657316341996193,
 0.0200123842805624,
 0.020429307594895363,
 0.029487943276762962,
 0.01567889377474785,
 0.023284610360860825,
 0.030827155336737633,
 0.03724526986479759,
 -0.01245720311999321,
 -0.014339681714773178,
 -0.05020783469080925,
 0.009159708395600319,
 -0.003430152777582407,
 0.003888138337060809,
 -0.0002218860317952931,
 -0.028376145288348198,
 0.017346592620015144,
 0.023966850712895393,
 0.05291152745485306,
 -0.017763517796993256,
 0.015236700884997845,
 -0.0033038121182471514,
 -0.005919066723436117,
 -0.004374550189822912,
 0.019001657143235207,
 -0.0017719297902658582,
 -0.016790693625807762,
 0.005116802640259266,
 -0.007984738796949387,
 -0.011503330431878567,
 -0.0035185914020985365,
 -0.03388460353016853,
 -0.026986395940184593,
 0.005028363782912493,
 -0.02523025870323181,
 -0.03229270875453949,
 -0.021124182268977165,
 0.013430027291178703,
 -0.013164712116122246,
 0.010707383044064045,
 -0.0272138100117445,
 0.01049892045557499,
 0.02409319207072258,
 0.020163992419838905,
 0.012640397064387798,
 -0.009551364928483963,
 -0.00644338084384799,
 -0.03633561357855797,
 -0.029260531067848206,
 -0.008142665028572083,
 0.015742063522338867,
 0.020366137847304344,
 0.0010249398183077574,
 0.010606310330331326,
 0.011225380003452301,
 0.007624667603522539,
 0.009747193194925785,
 -0.03181261569261551,
 -0.012804640457034111,
 0.004769365303218365,
 0.00830690748989582,
 -0.020947305485606194,
 0.022842416539788246,
 -0.011743377894163132,
 -0.04894442856311798,
 0.01647484116256237,
 0.009494511410593987,
 0.011143258772790432,
 0.014150169678032398,
 0.01977233588695526,
 0.018963754177093506,
 -0.0009965130593627691,
 0.006702379789203405,
 -0.002929527312517166,
 0.011452794075012207,
 -0.028856240212917328,
 0.004219783004373312,
 -0.022867685183882713,
 0.03294968232512474,
 0.05508458986878395,
 -0.022349689155817032,
 -0.006557087879627943,
 -0.024901771917939186,
 0.04502786323428154,
 -0.017447665333747864,
 -0.021819056943058968,
 0.02011345513164997,
 -0.040681738406419754,
 -0.0010296775726601481,
 0.022943489253520966,
 0.019671263173222542,
 -0.0050346809439361095,
 0.021263157948851585,
 -0.0057074460200965405,
 -0.012090815231204033,
 -0.0011962894350290298,
 -0.012292960658669472,
 0.029159458354115486,
 -0.0006060410523787141,
 0.02031560055911541,
 -0.01542621199041605,
 0.015502016991376877,
 -0.044118206948041916,
 -0.007409888319671154,
 0.022589735686779022,
 0.007814179174602032,
 -0.009412390179932117,
 -0.0090586356818676,
 0.02271607704460621,
 -0.028224535286426544,
 -0.004188197664916515,
 0.002675266470760107,
 -0.015224066562950611,
 0.019949212670326233,
 -0.01637376844882965,
 -0.0326717309653759,
 0.011073771864175797,
 0.012103448621928692,
 -0.02059355191886425,
 -0.00830690748989582,
 0.014604996889829636,
 -0.015565186738967896,
 0.024762798100709915,
 0.014402851462364197,
 -0.03178734704852104,
 0.020429307594895363,
 0.004017637576907873,
 0.037624292075634,
 0.00696769542992115,
 0.057813551276922226,
 0.016310598701238632,
 -0.032722268253564835,
 0.004052381496876478,
 0.0035312254913151264,
 0.006816086359322071,
 0.00017786415992304683,
 -0.030726084485650063,
 -0.01891321875154972,
 -0.020126089453697205,
 0.009715607389807701,
 -0.013948025181889534,
 -0.011332769878208637,
 -0.00497782789170742,
 0.0064812833443284035,
 0.03570391237735748,
 -0.013177345506846905,
 -0.002182537456974387,
 -0.015552553348243237,
 -0.013442661613225937,
 0.002740016207098961,
 -0.017182350158691406,
 0.0411112979054451,
 -0.008654344826936722,
 -0.01685386337339878,
 0.041237637400627136,
 0.006301247514784336,
 -0.036840979009866714,
 0.003932357300072908,
 0.009980923496186733,
 -0.005675860680639744,
 0.02089676819741726,
 0.06493917107582092,
 0.01785195618867874,
 -0.009153391234576702,
 0.007460424676537514,
 -0.029083652421832085,
 -0.014200706034898758,
 0.035602837800979614,
 -0.018989022821187973,
 -0.017750883474946022,
 0.022021202370524406,
 0.03292441368103027,
 -0.005682177841663361,
 0.017359226942062378,
 -0.01637376844882965,
 -0.00912812352180481,
 -0.016007380560040474,
 -0.016121085733175278,
 0.027744440361857414,
 -0.02701166458427906,
 0.0036607247311621904,
 -0.02069462463259697,
 -0.024990210309624672,
 0.02817399986088276,
 0.03666410222649574,
 -0.04495205730199814,
 0.005634800065308809,
 -0.03342977538704872,
 -0.020568283274769783,
 0.033859334886074066,
 -0.0026547361630946398,
 0.010113581083714962,
 0.014781873673200607,
 0.007371985819190741,
 -0.004090283531695604,
 0.038255997002124786,
 0.03350558131933212,
 -0.0020656720735132694,
 0.016209525987505913,
 0.033859334886074066,
 -0.0010202019475400448,
 -0.010922162793576717,
 -0.001951965386979282,
 0.013910122215747833,
 -0.00596328591927886,
 -0.05569102615118027,
 -0.0019803920295089483,
 -0.021010475233197212,
 0.00425452645868063,
 -0.0124256182461977,
 0.003106404561549425,
 -0.03075135126709938,
 -0.022905588150024414,
 0.003079557092860341,
 0.01771298050880432,
 0.004879913758486509,
 0.01921643689274788,
 0.011762328445911407,
 -0.006784501019865274,
 -0.025520842522382736,
 -0.03901404142379761,
 0.033202365040779114,
 0.03651249408721924,
 -0.01949438638985157,
 0.056853361427783966,
 -0.015363042242825031,
 0.015312505885958672,
 0.0073530348017811775,
 -0.015186164528131485,
 -0.004355599172413349,
 0.008591175079345703,
 0.00958926696330309,
 0.011534915305674076,
 0.020985208451747894,
 -0.01925433799624443,
 -0.02549557387828827,
 0.026127278804779053,
 -0.0447499118745327,
 0.01547674834728241,
 -0.0005452395416796207,
 0.034920599311590195,
 -0.0223370548337698,
 0.030245987698435783,
 0.04232417047023773,
 0.015742063522338867,
 -0.007100353017449379,
 0.018850047141313553,
 -0.019418582320213318,
 -0.021528473123908043,
 -0.018925853073596954,
 0.025242893025279045,
 0.018951119855046272,
 0.031105106696486473,
 0.03782643750309944,
 0.014213340356945992,
 -0.0230319295078516,
 0.0040207961574196815,
 0.0049367668107151985,
 -0.013733245432376862,
 -0.010170434601604939,
 0.03380879759788513,
 0.008660661987960339,
 0.02573562227189541,
 -0.01134540420025587,
 -0.025861961767077446,
 0.049980420619249344,
 -0.005249460227787495,
 0.010890576988458633,
 0.03858448192477226,
 -0.014743971638381481,
 -0.011263282969594002,
 -0.022589735686779022,
 -0.003392250509932637,
 -0.02824980393052101,
 0.008869124576449394,
 -0.00402395473793149,
 0.00573271419852972,
 0.0229561235755682,
 0.0387108214199543,
 -0.022084372118115425,
 0.003357506822794676,
 0.008534321561455727,
 -0.007409888319671154,
 0.027163272723555565,
 -0.01154754962772131,
 0.007270913105458021,
 0.016045281663537025,
 0.025647183880209923,
 -0.014263876713812351,
 0.03254539147019386,
 0.025242893025279045,
 0.03461737930774689,
 0.014478656463325024,
 -0.036461956799030304,
 0.00034664757549762726,
 0.0006245973636396229,
 0.026102010160684586,
 -0.013316321186721325,
 -0.012261374853551388,
 0.004469305742532015,
 -0.011768645606935024,
 0.02317090332508087,
 0.02327197603881359,
 -0.011206429451704025,
 -0.01527460291981697,
 -0.00888807512819767,
 -0.02693585865199566,
 0.017308689653873444,
 0.006279137916862965,
 ...]

Next, I can apply this function to each of the notes stored in the notes column of the notes_df DataFrame, and save it in a new column of the DataFrame called embedding:

notes_df['embedding'] = notes_df['notes'].apply(lambda x: get_embedding(x))

Below, I show the first 5 rows. You can see that there is a new column containing a collection of 1,536-length lists corresponding to the note’s embedding.

notes_df.head(5)
notes embedding
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms. [0.007875490933656693, -0.0006737920339219272, -0.048676371574401855, 0.016742711886763573, -0.01964789256453514, 0.007379627320915461, -0.0036548112984746695, 0.01582098752260208, -0.00981811247766018, 0.05731024220585823, -0.05661020055413246, 0.05338999629020691, -0.0089722266420722, 0.005787027534097433, 0.009333916008472443, 0.002053461503237486, 0.020791297778487206, -0.013055814430117607, -0.03626226261258125, 0.06151050329208374, -0.0013227908639237285, 0.06081046164035797, 0.021164653822779655, 0.0037043977063149214, -0.004646540153771639, 0.009497258812189102, 0.006889596581459045, -0.02360313944518566, 0.04398607835173607, -0.06272391229867935, 0.02380148507654667, -0.01809612847864628, 0.003272704314440489, -0.017582762986421585, 0.030848590657114983, 0.023124776780605316, 0.022611411288380623, -0.046296220272779465, -0.006352896336466074, -0.006597911473363638, -0.023031437769532204, -0.01648602820932865, -0.01208742056041956, -0.006551241967827082, -0.030731918290257454, 0.0008546366589143872, 0.04699626564979553, -0.018119463697075844, 0.004751546308398247, -0.02417484112083912, 0.010763171128928661, -0.026694998145103455, 0.006452069152146578, 0.008348020724952221, 0.010535657405853271, 0.02921515703201294, 0.02755838632583618, 0.012624121271073818, -0.013405836187303066, 0.0015488466015085578, 0.0514298751950264, -0.03217867389321327, 0.02334645763039589, 0.013347499072551727, -0.004766130819916725, -0.013499175198376179, 0.03761567920446396, 0.01333583239465952, 0.026461651548743248, 0.010774838738143444, 0.006154550705105066, -0.028328433632850647, 0.025574930012226105, 0.003345625475049019, -0.025994954630732536, 0.04412608593702316, 0.019204530864953995, -0.041139233857393265, 0.024034833535552025, 0.014700917527079582, 0.0011288204696029425, -0.012075753882527351, 0.043052684515714645, -0.0011660102754831314, -0.023159777745604515, -0.05987706780433655, -0.033765438944101334, -0.023778149858117104, 0.027441712096333504, -0.031735312193632126, 0.031525298953056335, 0.03994916006922722, 0.006761255208402872, -0.028771795332431793, -0.013347499072551727, -0.02634497731924057, 0.02550492435693741, 0.02359147183597088, -0.03294872120022774, 0.03000853955745697, ...]
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause. [-0.0345405712723732, -0.010776856914162636, -0.010913430713117123, 0.01344623975455761, -0.0371975414454937, 0.01843736506998539, 0.022373152896761894, -0.01380629651248455, 0.06128406524658203, -0.0055250017903745174, -0.02246006391942501, 0.03488821163773537, -0.060837097465991974, 0.019294051453471184, -0.005022164434194565, 0.017419274896383286, 0.014824386686086655, -0.016761241480708122, -0.030120572075247765, 0.0056708864867687225, 0.036700911819934845, 0.03622911125421524, -0.006679664831608534, -0.0573606938123703, 0.024992873892188072, 0.01998933218419552, 0.030319223180413246, -0.023142928257584572, 0.029872257262468338, -0.03113866224884987, 0.004680731799453497, -0.020436298102140427, 0.014042195864021778, -0.020846018567681313, -0.02521635591983795, 0.04705563187599182, 0.04226315766572952, 0.006909355986863375, -0.010733402334153652, -0.015283768996596336, -0.006704496685415506, -0.02394995093345642, 0.01163354329764843, -0.01688539795577526, -0.030865514650940895, -0.037445854395627975, -0.023502985015511513, -0.040425632148981094, 0.07345148175954819, 0.03406877443194389, -0.022149669006466866, 0.016016297042369843, 0.0220627598464489, -0.022075176239013672, 0.038935743272304535, -0.040624283254146576, 0.05368563532829285, 0.010013289749622345, -0.003945099655538797, 0.031188324093818665, 0.03250439092516899, 0.03357214480638504, 0.04186585545539856, 0.007896406576037407, 0.0016187013825401664, -0.0722595751285553, 0.01103137992322445, 0.01600388064980507, 0.008777923882007599, 0.043728217482566833, -0.005543625447899103, 0.010950678028166294, 0.005773316603153944, 0.05368563532829285, 0.01149076223373413, 0.024744559079408646, 0.006235802546143532, -0.028754839673638344, -0.00626994576305151, 0.009876716881990433, -0.02686764858663082, 0.006273049861192703, 0.009839469566941261, 0.046385183930397034, 0.0031737720128148794, -0.02080877125263214, -0.047105297446250916, -0.06217799708247185, 0.01106862723827362, -0.05676473677158356, 0.021963434293866158, -0.0022767353802919388, -0.00857306458055973, 0.015159611590206623, 0.005546729080379009, -0.050954174250364304, -0.0011461274698376656, 0.02709113247692585, -0.00894553679972887, 0.04352956265211105, ...]
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs. [-0.017208361998200417, 0.006888406351208687, -0.005633207969367504, 0.04822390526533127, -0.014728333801031113, 0.025873279199004173, -0.03370814397931099, -0.0009527865331619978, 0.03423452004790306, -0.010254159569740295, -0.04486321285367012, -0.008685161359608173, 0.01806878112256527, 0.019941454753279686, 0.028039507567882538, 0.016347944736480713, 0.0295174028724432, -0.05652441084384918, -0.020892975851893425, 0.009252025745809078, 0.053163718432188034, 0.015467280521988869, 0.02342361770570278, -0.0332222618162632, 0.04615890234708786, -0.051422636955976486, 0.01870650239288807, -0.00034986119135282934, -0.023038960993289948, -0.024010727182030678, 0.047697532922029495, -0.0066960775293409824, 0.03565167635679245, -0.026075730100274086, -0.035894621163606644, 0.06778070330619812, 0.03374863415956497, 0.0055269212462008, -0.015720345079898834, -0.02858612686395645, 0.009727786295115948, 0.024597834795713425, 0.019010178744792938, -0.020235009491443634, -0.007941152900457382, 0.025245679542422295, 0.008472587913274765, -0.02548862062394619, 0.02113591879606247, 0.0035783271305263042, 0.0027204395737499, 0.03320201858878136, 0.021115673705935478, -0.01451575942337513, 0.006088723428547382, -0.05980411916971207, -0.02872784249484539, -0.0322909876704216, 0.03806084766983986, 0.025144454091787338, 0.004529848229140043, 0.03957923501729965, 0.06259794533252716, 0.05907529592514038, -0.010821023024618626, -0.07883454114198685, 0.05713176354765892, 0.040915410965681076, 0.044458311051130295, -0.02583278901875019, 0.013078355230391026, 0.02773583121597767, -0.010238975286483765, 0.02148008532822132, 0.07474502921104431, 0.00839666835963726, 0.08215474337339401, -0.026865290477871895, 0.010861513204872608, -0.0009135616128332913, 0.004243885632604361, 0.019081037491559982, 0.005435817874968052, 0.03941727057099342, 0.014819436706602573, 0.02615671046078205, -0.0274321548640728, -0.034760892391204834, 0.02696651592850685, 0.00028643698897212744, 0.001967573771253228, 0.011448622681200504, 0.04247428849339485, -0.0023927215952426195, -0.006756812799721956, -0.03441672399640083, 0.001920756883919239, 0.03672466799616814, -0.031015543267130852, 0.02471930719912052, ...]
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise. [-0.01908160373568535, -0.007425420451909304, 0.017478106543421745, 0.0633997693657875, -0.03739846125245094, 0.0564923994243145, -0.014924848452210426, 0.03949534147977829, -0.022670967504382133, -0.008227168582379818, 0.0023235275875777006, 0.01861288957297802, -0.05338408425450325, 0.021585524082183838, 0.032809995114803314, 0.0043664430268108845, 0.05175592005252838, -0.01282796822488308, -0.01846487447619438, 0.010040352120995522, -0.01032404787838459, -0.01255044061690569, 0.03727511689066887, -0.03327871114015579, -0.01349403616040945, -0.02069743350148201, 0.026494689285755157, -0.05005374550819397, 0.042801011353731155, -0.04746348410844803, 0.04080280661582947, -0.02429913356900215, 0.025137884542346, 0.0548148974776268, -0.005189776886254549, -0.0036479535046964884, 0.0019658245146274567, -0.0015927032800391316, 0.022905325517058372, 0.0035554443020373583, -0.0008025189745239913, -0.028912268579006195, -0.020956460386514664, 0.0659160241484642, -0.03490687534213066, 0.02893693745136261, 0.017613787204027176, -0.03764515370130539, 0.0155539121478796, 0.02130182832479477, -0.019982028752565384, 0.010428891517221928, -0.023115012794733047, 0.019328294321894646, 0.03290867432951927, 0.015183874405920506, 0.03014572709798813, 0.020487746223807335, -0.007838629186153412, 0.031132493168115616, 0.014184772968292236, -0.013975084759294987, 0.022300930693745613, 0.025729944929480553, -0.04578598216176033, -0.02073443867266178, 0.015072863548994064, -0.011446495540440083, 0.03140385448932648, 0.021425174549221992, -0.014949517324566841, -0.030022380873560905, -0.030269071459770203, 0.05358143895864487, -0.0035708623472601175, -0.022892991080880165, -0.005436468403786421, -0.03009638749063015, -0.003157653845846653, 0.055308278650045395, -0.029603004455566406, -0.02204190380871296, -0.009065920487046242, 0.04381244629621506, 0.009756657294929028, -0.06902433931827545, 0.0014246446080505848, 9.824304288486019e-05, -0.005482723005115986, -0.08269105851650238, 0.0053686280734837055, 0.005710912868380547, -0.0006240529473870993, 0.004471287131309509, -0.03041708655655384, 0.003200824838131666, -0.04183891415596008, 0.03251396492123604, -0.0017438019858673215, -0.0027629469987004995, ...]
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated. [-0.010823288932442665, 0.03450469672679901, -0.0003536372387316078, 0.01706509292125702, -0.033481039106845856, 0.030859481543302536, -0.00017135703819803894, 0.05228135362267494, -0.010186624713242054, 0.017614372074604034, 0.003529740497469902, 0.021334487944841385, -0.02444290742278099, 0.054828010499477386, -0.012496092356741428, -0.005037136375904083, 0.008351534605026245, -0.02938641607761383, -0.005436611827462912, 0.02893700636923313, 0.03158353269100189, 0.06366640329360962, 0.05123273283243179, 0.015329872258007526, 0.0488358773291111, 0.01000561285763979, 0.010242801159620285, -0.041695255786180496, 0.03263215348124504, -0.022308209910988808, 0.02629048004746437, -0.00510579627007246, -0.018038814887404442, 0.03467946499586105, -0.005889142397791147, 0.022645266726613045, 0.025504013523459435, -0.004253789782524109, -0.013145240023732185, -0.0078896414488554, 0.016728036105632782, -0.020210962742567062, 0.016815422102808952, 0.01926220953464508, -0.015641963109374046, -0.046139419078826904, 0.02104736492037773, -0.03822481259703636, 0.051582273095846176, 0.006853501312434673, -0.039173565804958344, 0.012858116999268532, 0.02134697139263153, 0.013245109468698502, 0.014643273316323757, -0.010473747737705708, 0.003486047964543104, -0.024005981162190437, 0.06111975014209747, 0.02057298831641674, 0.052980437874794006, -0.003170836716890335, 0.02688969485461712, 0.030260268598794937, -0.03732599318027496, -0.037800367921590805, 0.03233254700899124, -0.04114597663283348, 0.06072027608752251, -0.013956675305962563, -0.019536849111318588, 0.005062103737145662, 0.04409210756421089, 0.004054052289575338, 0.008794702589511871, -0.013170207850635052, 0.05472814291715622, -0.06970847398042679, -0.025416627526283264, 0.05397912487387657, -0.0226327832788229, 0.025591399520635605, 0.008863362483680248, 0.04014728590846062, -0.012508576735854149, -0.05412892997264862, -0.029186679050326347, -0.03198300674557686, -0.015254970639944077, -0.03430495783686638, -0.024230685085058212, 0.012764490209519863, 0.010411330498754978, 0.012957986444234848, -0.016128823161125183, -0.012021715752780437, 0.015854183584451675, 0.00357655412517488, -0.015666930004954338, 0.02527930773794651, ...]

Since I will want to compute the cosine distance to the “target” (“Patient presenting with ongoing chronic condition defined as ongoing for more than one month”), I will also need to compute the embedding of the label.

target = 'Patient presenting with ongoing chronic condition defined as ongoing for more than one month'
# Compute embedding of the target
target_embedding = get_embedding(target)

Computing the cosine distance from the target

Now that I have all of my embeddings, I need to compute the cosine distance from each doctor’s note’s embedding to this target embedding. Below, I compute and save this cosine distance in a new column of notes_df called cosine_dist (note that a lower cosine_dist value corresponds to a doctor’s note that is more similar to the target: “Patient presenting with ongoing chronic condition defined as ongoing for more than one month”).

notes_df['cosine_dist'] = notes_df.embedding.apply(lambda x: cosine(x, target_embedding))

You can see that the final column in notes_df is now the cosine distance:

notes_df.head(5)
notes embedding cosine_dist
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms. [0.007875490933656693, -0.0006737920339219272, -0.048676371574401855, 0.016742711886763573, -0.01964789256453514, 0.007379627320915461, -0.0036548112984746695, 0.01582098752260208, -0.00981811247766018, 0.05731024220585823, -0.05661020055413246, 0.05338999629020691, -0.0089722266420722, 0.005787027534097433, 0.009333916008472443, 0.002053461503237486, 0.020791297778487206, -0.013055814430117607, -0.03626226261258125, 0.06151050329208374, -0.0013227908639237285, 0.06081046164035797, 0.021164653822779655, 0.0037043977063149214, -0.004646540153771639, 0.009497258812189102, 0.006889596581459045, -0.02360313944518566, 0.04398607835173607, -0.06272391229867935, 0.02380148507654667, -0.01809612847864628, 0.003272704314440489, -0.017582762986421585, 0.030848590657114983, 0.023124776780605316, 0.022611411288380623, -0.046296220272779465, -0.006352896336466074, -0.006597911473363638, -0.023031437769532204, -0.01648602820932865, -0.01208742056041956, -0.006551241967827082, -0.030731918290257454, 0.0008546366589143872, 0.04699626564979553, -0.018119463697075844, 0.004751546308398247, -0.02417484112083912, 0.010763171128928661, -0.026694998145103455, 0.006452069152146578, 0.008348020724952221, 0.010535657405853271, 0.02921515703201294, 0.02755838632583618, 0.012624121271073818, -0.013405836187303066, 0.0015488466015085578, 0.0514298751950264, -0.03217867389321327, 0.02334645763039589, 0.013347499072551727, -0.004766130819916725, -0.013499175198376179, 0.03761567920446396, 0.01333583239465952, 0.026461651548743248, 0.010774838738143444, 0.006154550705105066, -0.028328433632850647, 0.025574930012226105, 0.003345625475049019, -0.025994954630732536, 0.04412608593702316, 0.019204530864953995, -0.041139233857393265, 0.024034833535552025, 0.014700917527079582, 0.0011288204696029425, -0.012075753882527351, 0.043052684515714645, -0.0011660102754831314, -0.023159777745604515, -0.05987706780433655, -0.033765438944101334, -0.023778149858117104, 0.027441712096333504, -0.031735312193632126, 0.031525298953056335, 0.03994916006922722, 0.006761255208402872, -0.028771795332431793, -0.013347499072551727, -0.02634497731924057, 0.02550492435693741, 0.02359147183597088, -0.03294872120022774, 0.03000853955745697, ...] 0.697187
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause. [-0.0345405712723732, -0.010776856914162636, -0.010913430713117123, 0.01344623975455761, -0.0371975414454937, 0.01843736506998539, 0.022373152896761894, -0.01380629651248455, 0.06128406524658203, -0.0055250017903745174, -0.02246006391942501, 0.03488821163773537, -0.060837097465991974, 0.019294051453471184, -0.005022164434194565, 0.017419274896383286, 0.014824386686086655, -0.016761241480708122, -0.030120572075247765, 0.0056708864867687225, 0.036700911819934845, 0.03622911125421524, -0.006679664831608534, -0.0573606938123703, 0.024992873892188072, 0.01998933218419552, 0.030319223180413246, -0.023142928257584572, 0.029872257262468338, -0.03113866224884987, 0.004680731799453497, -0.020436298102140427, 0.014042195864021778, -0.020846018567681313, -0.02521635591983795, 0.04705563187599182, 0.04226315766572952, 0.006909355986863375, -0.010733402334153652, -0.015283768996596336, -0.006704496685415506, -0.02394995093345642, 0.01163354329764843, -0.01688539795577526, -0.030865514650940895, -0.037445854395627975, -0.023502985015511513, -0.040425632148981094, 0.07345148175954819, 0.03406877443194389, -0.022149669006466866, 0.016016297042369843, 0.0220627598464489, -0.022075176239013672, 0.038935743272304535, -0.040624283254146576, 0.05368563532829285, 0.010013289749622345, -0.003945099655538797, 0.031188324093818665, 0.03250439092516899, 0.03357214480638504, 0.04186585545539856, 0.007896406576037407, 0.0016187013825401664, -0.0722595751285553, 0.01103137992322445, 0.01600388064980507, 0.008777923882007599, 0.043728217482566833, -0.005543625447899103, 0.010950678028166294, 0.005773316603153944, 0.05368563532829285, 0.01149076223373413, 0.024744559079408646, 0.006235802546143532, -0.028754839673638344, -0.00626994576305151, 0.009876716881990433, -0.02686764858663082, 0.006273049861192703, 0.009839469566941261, 0.046385183930397034, 0.0031737720128148794, -0.02080877125263214, -0.047105297446250916, -0.06217799708247185, 0.01106862723827362, -0.05676473677158356, 0.021963434293866158, -0.0022767353802919388, -0.00857306458055973, 0.015159611590206623, 0.005546729080379009, -0.050954174250364304, -0.0011461274698376656, 0.02709113247692585, -0.00894553679972887, 0.04352956265211105, ...] 0.583531
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs. [-0.017208361998200417, 0.006888406351208687, -0.005633207969367504, 0.04822390526533127, -0.014728333801031113, 0.025873279199004173, -0.03370814397931099, -0.0009527865331619978, 0.03423452004790306, -0.010254159569740295, -0.04486321285367012, -0.008685161359608173, 0.01806878112256527, 0.019941454753279686, 0.028039507567882538, 0.016347944736480713, 0.0295174028724432, -0.05652441084384918, -0.020892975851893425, 0.009252025745809078, 0.053163718432188034, 0.015467280521988869, 0.02342361770570278, -0.0332222618162632, 0.04615890234708786, -0.051422636955976486, 0.01870650239288807, -0.00034986119135282934, -0.023038960993289948, -0.024010727182030678, 0.047697532922029495, -0.0066960775293409824, 0.03565167635679245, -0.026075730100274086, -0.035894621163606644, 0.06778070330619812, 0.03374863415956497, 0.0055269212462008, -0.015720345079898834, -0.02858612686395645, 0.009727786295115948, 0.024597834795713425, 0.019010178744792938, -0.020235009491443634, -0.007941152900457382, 0.025245679542422295, 0.008472587913274765, -0.02548862062394619, 0.02113591879606247, 0.0035783271305263042, 0.0027204395737499, 0.03320201858878136, 0.021115673705935478, -0.01451575942337513, 0.006088723428547382, -0.05980411916971207, -0.02872784249484539, -0.0322909876704216, 0.03806084766983986, 0.025144454091787338, 0.004529848229140043, 0.03957923501729965, 0.06259794533252716, 0.05907529592514038, -0.010821023024618626, -0.07883454114198685, 0.05713176354765892, 0.040915410965681076, 0.044458311051130295, -0.02583278901875019, 0.013078355230391026, 0.02773583121597767, -0.010238975286483765, 0.02148008532822132, 0.07474502921104431, 0.00839666835963726, 0.08215474337339401, -0.026865290477871895, 0.010861513204872608, -0.0009135616128332913, 0.004243885632604361, 0.019081037491559982, 0.005435817874968052, 0.03941727057099342, 0.014819436706602573, 0.02615671046078205, -0.0274321548640728, -0.034760892391204834, 0.02696651592850685, 0.00028643698897212744, 0.001967573771253228, 0.011448622681200504, 0.04247428849339485, -0.0023927215952426195, -0.006756812799721956, -0.03441672399640083, 0.001920756883919239, 0.03672466799616814, -0.031015543267130852, 0.02471930719912052, ...] 0.748090
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise. [-0.01908160373568535, -0.007425420451909304, 0.017478106543421745, 0.0633997693657875, -0.03739846125245094, 0.0564923994243145, -0.014924848452210426, 0.03949534147977829, -0.022670967504382133, -0.008227168582379818, 0.0023235275875777006, 0.01861288957297802, -0.05338408425450325, 0.021585524082183838, 0.032809995114803314, 0.0043664430268108845, 0.05175592005252838, -0.01282796822488308, -0.01846487447619438, 0.010040352120995522, -0.01032404787838459, -0.01255044061690569, 0.03727511689066887, -0.03327871114015579, -0.01349403616040945, -0.02069743350148201, 0.026494689285755157, -0.05005374550819397, 0.042801011353731155, -0.04746348410844803, 0.04080280661582947, -0.02429913356900215, 0.025137884542346, 0.0548148974776268, -0.005189776886254549, -0.0036479535046964884, 0.0019658245146274567, -0.0015927032800391316, 0.022905325517058372, 0.0035554443020373583, -0.0008025189745239913, -0.028912268579006195, -0.020956460386514664, 0.0659160241484642, -0.03490687534213066, 0.02893693745136261, 0.017613787204027176, -0.03764515370130539, 0.0155539121478796, 0.02130182832479477, -0.019982028752565384, 0.010428891517221928, -0.023115012794733047, 0.019328294321894646, 0.03290867432951927, 0.015183874405920506, 0.03014572709798813, 0.020487746223807335, -0.007838629186153412, 0.031132493168115616, 0.014184772968292236, -0.013975084759294987, 0.022300930693745613, 0.025729944929480553, -0.04578598216176033, -0.02073443867266178, 0.015072863548994064, -0.011446495540440083, 0.03140385448932648, 0.021425174549221992, -0.014949517324566841, -0.030022380873560905, -0.030269071459770203, 0.05358143895864487, -0.0035708623472601175, -0.022892991080880165, -0.005436468403786421, -0.03009638749063015, -0.003157653845846653, 0.055308278650045395, -0.029603004455566406, -0.02204190380871296, -0.009065920487046242, 0.04381244629621506, 0.009756657294929028, -0.06902433931827545, 0.0014246446080505848, 9.824304288486019e-05, -0.005482723005115986, -0.08269105851650238, 0.0053686280734837055, 0.005710912868380547, -0.0006240529473870993, 0.004471287131309509, -0.03041708655655384, 0.003200824838131666, -0.04183891415596008, 0.03251396492123604, -0.0017438019858673215, -0.0027629469987004995, ...] 0.644881
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated. [-0.010823288932442665, 0.03450469672679901, -0.0003536372387316078, 0.01706509292125702, -0.033481039106845856, 0.030859481543302536, -0.00017135703819803894, 0.05228135362267494, -0.010186624713242054, 0.017614372074604034, 0.003529740497469902, 0.021334487944841385, -0.02444290742278099, 0.054828010499477386, -0.012496092356741428, -0.005037136375904083, 0.008351534605026245, -0.02938641607761383, -0.005436611827462912, 0.02893700636923313, 0.03158353269100189, 0.06366640329360962, 0.05123273283243179, 0.015329872258007526, 0.0488358773291111, 0.01000561285763979, 0.010242801159620285, -0.041695255786180496, 0.03263215348124504, -0.022308209910988808, 0.02629048004746437, -0.00510579627007246, -0.018038814887404442, 0.03467946499586105, -0.005889142397791147, 0.022645266726613045, 0.025504013523459435, -0.004253789782524109, -0.013145240023732185, -0.0078896414488554, 0.016728036105632782, -0.020210962742567062, 0.016815422102808952, 0.01926220953464508, -0.015641963109374046, -0.046139419078826904, 0.02104736492037773, -0.03822481259703636, 0.051582273095846176, 0.006853501312434673, -0.039173565804958344, 0.012858116999268532, 0.02134697139263153, 0.013245109468698502, 0.014643273316323757, -0.010473747737705708, 0.003486047964543104, -0.024005981162190437, 0.06111975014209747, 0.02057298831641674, 0.052980437874794006, -0.003170836716890335, 0.02688969485461712, 0.030260268598794937, -0.03732599318027496, -0.037800367921590805, 0.03233254700899124, -0.04114597663283348, 0.06072027608752251, -0.013956675305962563, -0.019536849111318588, 0.005062103737145662, 0.04409210756421089, 0.004054052289575338, 0.008794702589511871, -0.013170207850635052, 0.05472814291715622, -0.06970847398042679, -0.025416627526283264, 0.05397912487387657, -0.0226327832788229, 0.025591399520635605, 0.008863362483680248, 0.04014728590846062, -0.012508576735854149, -0.05412892997264862, -0.029186679050326347, -0.03198300674557686, -0.015254970639944077, -0.03430495783686638, -0.024230685085058212, 0.012764490209519863, 0.010411330498754978, 0.012957986444234848, -0.016128823161125183, -0.012021715752780437, 0.015854183584451675, 0.00357655412517488, -0.015666930004954338, 0.02527930773794651, ...] 0.688291

And that’s it! Let’s take a look at the notes in order of smallest to largest cosine_dist.

Ideally, I will see that the notes at the top of the rearranged DataFrame correspond to patients with chronic conditions, and the notes at the bottom correspond to patients with acute conditions.

The top 5 notes are:

pd.set_option('display.max_colwidth', None)
notes_df.sort_values('cosine_dist')[['notes', 'cosine_dist']].head(5)
notes cosine_dist
43 Patient with chronic bronchitis showed increasingly frequent flare-ups. Reinforced importance of smoking cessation and upgraded prescribed inhalers. 0.483188
19 Patient visited for follow-up on hypertension medication. Blood pressure readings stable. Emphasized continued adherence to treatment. 0.547638
27 Patient reported persistent cough and difficulty breathing. Referred for radiological imaging to rule out pneumonia or tuberculosis. 0.550634
22 Patient with a history of heart disease complained of palpitations. Recommended further cardiology work-up and emphasized continued adherence to treatment. 0.558663
31 Chronic back pain patient showed improvement after physiotherapy. Recommended continuing sessions and daily exercises at home. 0.565519

Those all seem like patients with chronic conditions to me!

And the bottom 5 notes are:

notes_df.sort_values('cosine_dist')[['notes', 'cosine_dist']].tail(5)
notes cosine_dist
39 Child patient presented with red, itchy eyes. Diagnosed as viral conjunctivitis. Instructed parents about proper eye care and hand hygiene. 0.743466
16 Patient suffered a minor burn in the kitchen. Prescribed topical ointment and advised on wound care. 0.745391
26 Patient presented with severe knee pain. Suspected ligament tear. Recommended MRI scan. 0.746433
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs. 0.748090
49 Patient presented with dislocated shoulder. After successful relocation, advised sling use and gentle mobility exercises for rehabilitation. 0.748190

These definitely seem like new/acute conditions!

If you’re interested in looking at all of the distances, see the following output (remember that the smaller the distance, the closer the note to the chronic condition target):

notes_df.sort_values('cosine_dist')[['notes', 'cosine_dist']]
notes cosine_dist
43 Patient with chronic bronchitis showed increasingly frequent flare-ups. Reinforced importance of smoking cessation and upgraded prescribed inhalers. 0.483188
19 Patient visited for follow-up on hypertension medication. Blood pressure readings stable. Emphasized continued adherence to treatment. 0.547638
27 Patient reported persistent cough and difficulty breathing. Referred for radiological imaging to rule out pneumonia or tuberculosis. 0.550634
22 Patient with a history of heart disease complained of palpitations. Recommended further cardiology work-up and emphasized continued adherence to treatment. 0.558663
31 Chronic back pain patient showed improvement after physiotherapy. Recommended continuing sessions and daily exercises at home. 0.565519
15 Asthmatic patient presented with increased shortness of breath. Stepped-up asthma control medication. 0.578419
11 Patient admitted for chronic fatigue. Bloodwork ordered to rule out anemia, thyroid disorders, and diabetes. Recommended nutritional counseling. 0.580687
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause. 0.583531
5 Patient came in for annual physical. All markers within normal limits. Encouraged to continue healthy lifestyle for prevention of non-communicable diseases. 0.592007
13 Patient managing COPD came in for routine check-up. Oxygen levels stable. Expressed importance of avoiding respiratory irritants. 0.597322
35 Patient presented with swollen (oedemic) ankles. Adjusted diuretic medication dosage and recommended follow-up visit next week. 0.598728
30 Elderly patient presented with signs of Parkinson's disease. Referred to a neurologist for specialized care and management. 0.601647
51 Patient complained of periodic blurry vision. Referred to ophthalmology for proper ocular examination and management. 0.607228
24 Patient managed for rheumatoid arthritis reported increased inflammation in joints. Adjusted medication and recommended physiotherapy for managing symptoms. 0.610967
28 Patient presented with severe migraine episodes. Prescribed stronger painkillers and referred to a neurologist for a complete neurological work-up. 0.613839
41 Patient presented with painful and swollen joints. Diagnosed as gout and recommended dietary adjustment along with pain management. 0.618245
25 Elderly patient with a history of diabetes complained of numbness in feet. Diagnosed peripheral neuropathy, referred to podiatrist for further management. 0.619704
37 Elderly patient exhibited signs of Alzheimer's. Informed family and referred to neurology specialist for comprehensive management. 0.619889
10 Patient presented with chest pain. Preliminary assessment ruled out heart conditions. Recommended gastroenterology consult to check for possible reflux disease. 0.625152
46 Patient with anxiety showed improvement with recent medication schedule. Encouraged continued therapy and stress management techniques. 0.626963
18 Patient reports persistent vertigo. Referred to an Ear, Nose, and Throat (ENT) specialist for detailed evaluation to rule out Meniere's disease. 0.628280
42 Patient reported sleep apnea symptoms. Advised to undergo a sleep study, prescribed initial treatment using CPAP device at night during sleep. 0.636255
48 Child patient presented with symptoms of ADHD. Referred to pediatric psychiatrist for cognitive and behavioral evaluation plus treatment. 0.636682
44 Patient with history of stroke presented with dizziness. Follow-up brain scan scheduled to rule out recurrence. 0.637395
32 Patient came in for weight management consultation. Diet plan adjusted and encouraged increased physical activity. 0.639904
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise. 0.644881
20 Historically anxious patient reported increased panic attacks recently. Referred to psychiatrist for therapeutic support, prescribed low-dose anxiolytic. 0.645849
45 Patient reported feeling low, indicating depression. Counseled and referred to mental health services for further treatment and support. 0.647704
12 Elderly patient with poor vision due to cataracts. Scheduled for ophthalmology consultation. 0.648220
38 Patient with history of irritable bowel syndrome showed signs of relief with diet modification. Reinforced importance of hydration and fiber intake. 0.657894
36 Teenage patient came in with signs of acne. Prescribed topical retinoid and advised consistent skin care routine. 0.658439
9 Patient returned for post-operative wound check. Incision healing well with no signs of infection. Advised continued wound care protocol and scheduled follow-up in a week. 0.660693
8 Arthritic patient reports increased joint pain. Advised gentle physical therapy and adjustment of pain medication. 0.661807
14 Patient presented with signs of depression. Recommended seeking a psychologist for cognitive behavioral therapy and prescribed low dose antidepressant. 0.666589
52 Elderly patient with hearing difficulty. Scheduled an appointment for audiology consultation and potential hearing aids. 0.667571
6 Routine consultation with patient managing type II diabetes. Blood glucose levels fairly controlled. Advised to keep monitoring sugar levels and follow diet regimen. 0.669839
21 Patient has concerning mole on the back. Scheduled for dermatology consult to rule out potential skin cancer. 0.675400
29 Patient suffering from high fevers and weight loss. Ordered comprehensive bloodwork to assess for possible infectious disease. 0.676904
53 Patient presented with signs of Bell's Palsy. Referred to neurologist and prescribed corticosteroid medication to reduce inflammation. 0.682327
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated. 0.688291
50 Pregnant patient at full term demonstrated normal signs. Advised to prepare for labor and familiarize with signs of onset. 0.696364
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms. 0.697187
33 Pregnant patient showed normal progression in second trimester. Advised rest, proper nutrition and regular exercise. 0.702353
47 Type 1 diabetic patient showed increased blood sugar levels. Insulin doses were adjusted, advised dietary check. 0.707895
40 Patient showed symptoms of urinary tract infection. Collected sample for testing and started preliminary antibiotic treatment. 0.710449
7 Patient presented with symptoms of influenza including fever, cough, and body aches. Prescribed antiviral medication and recommended rest and plenty of fluids. 0.715153
23 Post-surgical patient reported redness and discomfort around stitches. Diagnosed with a mild post-operative infection. Prescribed antibiotics and wound care guidance. 0.726964
17 Rash observed on patient's arm. Allergenic contact dermatitis suspected. Prescribed topical steroid and recommended patch test. 0.730986
34 Postnatal patient showed healthy recovery from C-section. Wound healing well without signs of infection. 0.742733
39 Child patient presented with red, itchy eyes. Diagnosed as viral conjunctivitis. Instructed parents about proper eye care and hand hygiene. 0.743466
16 Patient suffered a minor burn in the kitchen. Prescribed topical ointment and advised on wound care. 0.745391
26 Patient presented with severe knee pain. Suspected ligament tear. Recommended MRI scan. 0.746433
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs. 0.748090
49 Patient presented with dislocated shoulder. After successful relocation, advised sling use and gentle mobility exercises for rehabilitation. 0.748190

It definitely seems like it did a pretty good job here (although to be fair, OpenAI did all the hard lifting!)

Zero-shot Binary classification

The above approach provides each doctor’s note with a score (cosine distance) relative to a target (“Patient presenting with ongoing chronic condition defined as ongoing for more than one month”). But if I wanted to use this to create a binary label, I would need to come up with a threshold so that everyone whose cosine distance to the target is below, say 0.6, is classified as “presenting with a chronic condition”, in which case, everyone whose cosine distance to the target is above 0.6 is classified as “presenting with an acute condition”.

That approach sounds okay, but it turns out that there’s a (hypothetically) better way.

Instead of just computing the distance to a single target, I can compute two distances for each doctor’s note:

  1. the distance to a positive label (corresponding to the original target: “Patient presenting with ongoing chronic condition defined as ongoing for more than one month”)

  2. the distance to a negative label (“Patient presenting with acute or new condition”)

Then, each note is classified based on which label text its embedding is closest to.

The code for doing this is very similar to the code above.

Fortunately, I don’t need to recompute the embeddings for the notes, but I do need to compute the embedding for the new negative label.

labels = [
  'Patient presenting with ongoing chronic condition defined as ongoing for more than one month',
  'Patient presenting with acute or new condition'
]
# Compute embedding of the labels
label_embeddings = [get_embedding(label) for label in labels]

Then I need to compute the difference between the cosine distance to the first label and the cosine distance to the second label (to determine which label the note is closest to).

# compute the difference in the cosine distances to the two labels for each note
notes_df['cosine_score_binary'] = [cosine(x, label_embeddings[1]) - cosine(x, label_embeddings[0]) for x in notes_df['embedding']]

I have saved this cosine_score_binary in a new column in the DataFrame:

notes_df[['notes', 'cosine_score_binary']].head()
notes cosine_score_binary
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms. -0.097242
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause. 0.061629
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs. -0.074738
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise. -0.009268
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated. -0.013830

A positive cosine_score_binary value means that the note is closer to the first label, “Patient presenting with ongoing chronic condition defined as ongoing for more than one month” (i.e., the distance to the second label is greater than the distance to the first label), and a negative value means that the note is closer to the second label, “Patient presenting with acute or new condition”.

I can use this to create a binary classification for each note.

notes_df['chronic_assessment'] = ['chronic' if score > 0 else 'acute' for score in notes_df['cosine_score_binary']]

Then I can view the results in descending order of the cosine_score_binary so that the first notes are those that are deemed most chronic and least acute:

notes_df[['notes', 'cosine_score_binary', 'chronic_assessment']].sort_values('cosine_score_binary', ascending=False)
notes cosine_score_binary chronic_assessment
31 Chronic back pain patient showed improvement after physiotherapy. Recommended continuing sessions and daily exercises at home. 0.110414 chronic
43 Patient with chronic bronchitis showed increasingly frequent flare-ups. Reinforced importance of smoking cessation and upgraded prescribed inhalers. 0.091766 chronic
19 Patient visited for follow-up on hypertension medication. Blood pressure readings stable. Emphasized continued adherence to treatment. 0.081715 chronic
1 Patient reports difficulty sleeping. Exhibiting signs of insomnia characterized by frequent night time awakenings. Recommended a sleep study to find underlying cause. 0.061629 chronic
6 Routine consultation with patient managing type II diabetes. Blood glucose levels fairly controlled. Advised to keep monitoring sugar levels and follow diet regimen. 0.055009 chronic
11 Patient admitted for chronic fatigue. Bloodwork ordered to rule out anemia, thyroid disorders, and diabetes. Recommended nutritional counseling. 0.053559 chronic
51 Patient complained of periodic blurry vision. Referred to ophthalmology for proper ocular examination and management. 0.049252 chronic
5 Patient came in for annual physical. All markers within normal limits. Encouraged to continue healthy lifestyle for prevention of non-communicable diseases. 0.045713 chronic
25 Elderly patient with a history of diabetes complained of numbness in feet. Diagnosed peripheral neuropathy, referred to podiatrist for further management. 0.037681 chronic
13 Patient managing COPD came in for routine check-up. Oxygen levels stable. Expressed importance of avoiding respiratory irritants. 0.035781 chronic
27 Patient reported persistent cough and difficulty breathing. Referred for radiological imaging to rule out pneumonia or tuberculosis. 0.030508 chronic
18 Patient reports persistent vertigo. Referred to an Ear, Nose, and Throat (ENT) specialist for detailed evaluation to rule out Meniere's disease. 0.028819 chronic
24 Patient managed for rheumatoid arthritis reported increased inflammation in joints. Adjusted medication and recommended physiotherapy for managing symptoms. 0.026801 chronic
30 Elderly patient presented with signs of Parkinson's disease. Referred to a neurologist for specialized care and management. 0.019938 chronic
9 Patient returned for post-operative wound check. Incision healing well with no signs of infection. Advised continued wound care protocol and scheduled follow-up in a week. 0.018720 chronic
42 Patient reported sleep apnea symptoms. Advised to undergo a sleep study, prescribed initial treatment using CPAP device at night during sleep. 0.014099 chronic
22 Patient with a history of heart disease complained of palpitations. Recommended further cardiology work-up and emphasized continued adherence to treatment. 0.009364 chronic
37 Elderly patient exhibited signs of Alzheimer's. Informed family and referred to neurology specialist for comprehensive management. 0.007949 chronic
15 Asthmatic patient presented with increased shortness of breath. Stepped-up asthma control medication. 0.002814 chronic
32 Patient came in for weight management consultation. Diet plan adjusted and encouraged increased physical activity. 0.001463 chronic
33 Pregnant patient showed normal progression in second trimester. Advised rest, proper nutrition and regular exercise. 0.000649 chronic
12 Elderly patient with poor vision due to cataracts. Scheduled for ophthalmology consultation. -0.000179 acute
36 Teenage patient came in with signs of acne. Prescribed topical retinoid and advised consistent skin care routine. -0.000709 acute
41 Patient presented with painful and swollen joints. Diagnosed as gout and recommended dietary adjustment along with pain management. -0.004671 acute
21 Patient has concerning mole on the back. Scheduled for dermatology consult to rule out potential skin cancer. -0.007328 acute
3 Hypertensive patient presented with higher blood pressure at 160/100. Adjusted medications and emphasized importance of diet and exercise. -0.009268 acute
38 Patient with history of irritable bowel syndrome showed signs of relief with diet modification. Reinforced importance of hydration and fiber intake. -0.011980 acute
4 Examined patient for frequent headaches. No alarming neurological signs detected. Recommended reducing caffeine intake and keeping hydrated. -0.013830 acute
28 Patient presented with severe migraine episodes. Prescribed stronger painkillers and referred to a neurologist for a complete neurological work-up. -0.015010 acute
52 Elderly patient with hearing difficulty. Scheduled an appointment for audiology consultation and potential hearing aids. -0.015633 acute
45 Patient reported feeling low, indicating depression. Counseled and referred to mental health services for further treatment and support. -0.018623 acute
35 Patient presented with swollen (oedemic) ankles. Adjusted diuretic medication dosage and recommended follow-up visit next week. -0.020730 acute
47 Type 1 diabetic patient showed increased blood sugar levels. Insulin doses were adjusted, advised dietary check. -0.020986 acute
17 Rash observed on patient's arm. Allergenic contact dermatitis suspected. Prescribed topical steroid and recommended patch test. -0.030286 acute
34 Postnatal patient showed healthy recovery from C-section. Wound healing well without signs of infection. -0.032995 acute
8 Arthritic patient reports increased joint pain. Advised gentle physical therapy and adjustment of pain medication. -0.033021 acute
48 Child patient presented with symptoms of ADHD. Referred to pediatric psychiatrist for cognitive and behavioral evaluation plus treatment. -0.038751 acute
53 Patient presented with signs of Bell's Palsy. Referred to neurologist and prescribed corticosteroid medication to reduce inflammation. -0.042732 acute
46 Patient with anxiety showed improvement with recent medication schedule. Encouraged continued therapy and stress management techniques. -0.048093 acute
44 Patient with history of stroke presented with dizziness. Follow-up brain scan scheduled to rule out recurrence. -0.053002 acute
14 Patient presented with signs of depression. Recommended seeking a psychologist for cognitive behavioral therapy and prescribed low dose antidepressant. -0.057786 acute
29 Patient suffering from high fevers and weight loss. Ordered comprehensive bloodwork to assess for possible infectious disease. -0.059315 acute
16 Patient suffered a minor burn in the kitchen. Prescribed topical ointment and advised on wound care. -0.062896 acute
23 Post-surgical patient reported redness and discomfort around stitches. Diagnosed with a mild post-operative infection. Prescribed antibiotics and wound care guidance. -0.068201 acute
2 Patient came in with complaints of stomachache. Upon examination, no obvious signs of appendicitis or gallstones were found. Recommended patient to undergo an ultrasound to assess internal organs. -0.074738 acute
39 Child patient presented with red, itchy eyes. Diagnosed as viral conjunctivitis. Instructed parents about proper eye care and hand hygiene. -0.077558 acute
10 Patient presented with chest pain. Preliminary assessment ruled out heart conditions. Recommended gastroenterology consult to check for possible reflux disease. -0.080826 acute
26 Patient presented with severe knee pain. Suspected ligament tear. Recommended MRI scan. -0.081825 acute
20 Historically anxious patient reported increased panic attacks recently. Referred to psychiatrist for therapeutic support, prescribed low-dose anxiolytic. -0.083702 acute
49 Patient presented with dislocated shoulder. After successful relocation, advised sling use and gentle mobility exercises for rehabilitation. -0.091579 acute
0 Patient presented signs of common cold; symptoms include a runny nose and mild fever. Advised rest, hydration, and over-the-counter medication to manage symptoms. -0.097242 acute
7 Patient presented with symptoms of influenza including fever, cough, and body aches. Prescribed antiviral medication and recommended rest and plenty of fluids. -0.100791 acute
40 Patient showed symptoms of urinary tract infection. Collected sample for testing and started preliminary antibiotic treatment. -0.101196 acute
50 Pregnant patient at full term demonstrated normal signs. Advised to prepare for labor and familiarize with signs of onset. -0.101702 acute

While the results aren’t perfect, they’re certainly pretty impressive! Especially given how little code we had to write to achieve it!

Hopefully, this tutorial has given you some ideas to try in your own work. But keep in mind that this is a rapidly evolving field, and the specific pieces of code used here may or may not work in the future as things continue to shift.