add method predict_code for prediction code by group

app.py

@@ -19,7 +19,7 @@ try:
 except Exception as e:
     print(f"ERROR: loading embedder failed with: {str(e)}")
 
-
+print('Loading classifiers of codes')
 classifiers_codes = {}
 try:
     for clf_name in os.listdir('classifiers/codes'):
@@ -28,10 +28,11 @@ try:
         with open('classifiers/codes/'+clf_name, 'rb') as f:
             model = pickle.load(f)
             classifiers_codes[clf_name.split('.')[0]] = model
-        print(f'INFO: classifier {clf_name} loaded')
+        print(f'INFO: codes classifier {clf_name} loaded')
 except Exception as e:
     print(f"ERROR: loading classifiers failed with: {str(e)}")
 
+print('Loading classifiers of groups')
 classifiers_groups = {}
 try:
     for clf_name in os.listdir('classifiers/groups'):
@@ -40,7 +41,21 @@ try:
         with open('classifiers/groups/'+clf_name, 'rb') as f:
             model = pickle.load(f)
             classifiers_groups[clf_name.split('.')[0]] = model
-        print(f'INFO: classifier {clf_name} loaded')
+        print(f'INFO: groups classifier {clf_name} loaded')
+except Exception as e:
+    print(f"ERROR: loading classifiers failed with: {str(e)}")
+
+print('Loading classifiers in groups')
+groups_models = {}
+try:
+    for clf_name in os.listdir('classifiers/codes_in_groups'):
+        if '.' == clf_name[0]:
+            continue
+        with open('classifiers/codes_in_groups/'+clf_name, 'rb') as f:
+            model = pickle.load(f)
+            group_name = clf_name.replace('_code_clf.pkl', '')
+            groups_models[group_name] = model
+        print(f'INFO: codes classifier for group {group_name} loaded')
 except Exception as e:
     print(f"ERROR: loading classifiers failed with: {str(e)}")
 
@@ -68,6 +83,17 @@ def classify_group(text, top_n):
         preds[clf_name] = clf_preds
     return preds
 
+def classify_code_by_group(text, group_name, top_n):
+    embed = [embedder(text)]
+    model = groups_models[group_name]
+    probs = model.predict_proba(embed)
+    best_n = np.flip(np.argsort(probs, axis=1,)[0,-top_n:])
+
+    top_n_preds = {str(model.classes_[i]): float(probs[0][i]) for i in best_n}
+    top_cls = model.classes_[best_n[0]]
+    all_codes_in_group = model.classes_
+    return top_cls, top_n_preds, all_codes_in_group
+
 def get_top_result(preds):
     total_scores = {}
     for clf_name, scores in preds.items():
@@ -97,7 +123,7 @@ def test():
     return {'response': data}
 
 @app.route("/predict", methods=['POST'])
-def read_root():
+def predict_api():
     data = request.json
     base64_bytes = str(data['textB64']).encode("ascii")
     sample_string_bytes = base64.b64decode(base64_bytes)
@@ -121,5 +147,28 @@ def read_root():
     }
     return result
 
+@app.route("/predict_code", methods=['POST'])
+def predict_code_api():
+    data = request.json
+    base64_bytes = str(data['textB64']).encode("ascii")
+    sample_string_bytes = base64.b64decode(base64_bytes)
+    text = sample_string_bytes.decode("ascii")
+    top_n = int(data['top_n'])
+    group_name = data['dx_group']
+
+    if top_n < 1:
+        return {'error': 'top_n should be geather than 0'}
+    if text.strip() == '':
+        return {'error': 'text is empty'}
+    if group_name not in groups_models:
+        return {'error': 'have no classifier for the group'}
+    
+    top_pred_code, pred_codes, all_codes_in_group = classify_code_by_group(text, group_name, top_n)
+    result = {
+        "icd10":
+            {'result': top_pred_code, 'details': pred_codes, 'all_codes': all_codes_in_group}
+    }
+    return result
+
 if __name__ == "__main__":
     app.run(host='0.0.0.0', port=7860)