To start the API as a docker container run docker-compose up
To start the API outside of the container, change directory into the api folder (cd api) then run python classificationAPI.py
API Call: curl -i -H "Content-Type: application/json" -X POST -d '{"claim_text":"Ringing in my ear"}' http://localhost:8000/api/v1.0/classification
Response: {"prediction":{"classification":"hearing loss","code":3140,"probability":89},"text":"Ringing in my ear"}
To run the web interface run python ClassificationSite.py and open up a browser in https://localhost:8000
This classifier is designed to accept a string of text containing a disability description from a 526 form and return the proper VA approved classification.
IMPORTANT: The model needs to be trained once before running predictor.py or predictorBulk.py to generate the pickled files for Count Vectorizer and the Logistic Regression model. Simply follow the steps below and things should work smoothly
-
cd into the preppingScripts folder. cd preppingScripts
-
Run dataCleanUp.py with with the name of the csv containing your dataset as well as the name you'd like to give the cleaned up version of your data. Important:Make sure you feed and save files in CSV format -Example: python dataCleanUp.py Large_Training.csv CleanData.csv
-
Run modelBuilder.py with the name of the file you just created in step (1). -Example: python modelBuilder.py ../data/CleanData.csv This will generate the models and automatically save them.
Run predictor.py followed by the string of text you would like to analyze. -Example: python predictor.py 'Ringing in my ear' -Output: {'ringing in my ear': ['hearing loss', 3140]}
Run predictorBulk followed by the name of the csv file containing the data. -Example: python predictorBulk.py NewData.csv
Due to size contrains models and datasets will be shared independently. Models need to be dropped inside the tM folder for the code to work.
Accuracy: 92%
Weighted Precision: 91%
Weighted Recall: 92%
Weighted F1 Score 91%
A file named testResults.csv. is included in the data folder. It contains the following fields:
index: index of test elements in larger set.
CLMANT_TXT: The text from the 526 form.
CNTNTN_CLSFCN_TXT: The classification assigned to it.
CNTNTN_CLSFCN_ID: The ID for the assigned classification.
newClass: the class that corresponds to CNTNTN_CLSFCN_TXT based on string similarity with the approved list of classifications.
predictedLabel: The classification assigned by the model.
predID: The ID corresponding to the the predicted classification.
correctPred: Boolean for filtering based on whether or not the model made the right selection.
Created by: Bennett Gebken
Retrained and packaged by: Nel Abdiel