Update src/api.py

src/api.py

@@ -1,36 +1,48 @@
-from fastapi import FastAPI
-from pydantic import BaseModel
-from transformers import BertTokenizer, BertForSequenceClassification
-import torch
-import pickle
-
-app = FastAPI()
-
-# Label encoder yüklənməsi
-with open("label_encoder.pkl", "rb") as f:
-    label_encoder = pickle.load(f)
-
-# Model və tokenizer yüklənməsi
-tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
-model = BertForSequenceClassification.from_pretrained('bert-base-uncased', num_labels=len(label_encoder.classes_))
-model.eval()
-
-# Request modeli
-class TextRequest(BaseModel):
-    text: str
-
-@app.get("/")
-def home():
-    return {"message": "Disease prediction API is running!"}
-
-@app.post("/predict")
-async def predict_endpoint(request: TextRequest):
-    # Tokenize giriş mətni
-    inputs = tokenizer(request.text, return_tensors="pt", truncation=True, padding=True, max_length=128)
-    with torch.no_grad():
-        outputs = model(**inputs)
-        probs = torch.nn.functional.softmax(outputs.logits, dim=-1).squeeze().tolist()
-
-    # Label-ları geri çevir
-    labels = label_encoder.classes_  # 'classes_' ilə etiketləri alırıq
-    return {"predictions": dict(zip(labels, probs))}
+# api.py
+from fastapi import FastAPI, Query
+import uvicorn
+import pickle, torch, random, os
+from transformers import BertTokenizer, BertForSequenceClassification
+from collections import defaultdict
+
+app = FastAPI()
+
+# Load model
+label_encoder = pickle.load(open("best_model/label_encoder.pkl", "rb"))
+tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
+model = BertForSequenceClassification.from_pretrained(
+    "best_model", num_labels=len(label_encoder.classes_)
+)
+model.eval()
+
+def predict_disease(symptoms_text):
+    symptoms = [s.strip() for s in symptoms_text.split(",") if s.strip()]
+    agg_probs = defaultdict(float)
+    for _ in range(10):
+        random.shuffle(symptoms)
+        inputs = tokenizer(", ".join(symptoms), return_tensors="pt",
+                           truncation=True, padding=True, max_length=128)
+        with torch.no_grad():
+            outputs = model(**inputs)
+            probs = torch.nn.functional.softmax(outputs.logits, dim=-1).squeeze()
+        for i, p in enumerate(probs):
+            agg_probs[i] += p.item()
+    for k in agg_probs:
+        agg_probs[k] /= 10
+    return sorted(
+        [{"disease": label_encoder.classes_[i], "probability": p}
+         for i, p in agg_probs.items()],
+        key=lambda x: x["probability"], reverse=True
+    )[:3]
+
+@app.get("/")
+def api(symptoms: str = Query(...)):
+    results = predict_disease(symptoms)
+    return {
+        "status": "success",
+        "input": symptoms,
+        "predictions": results
+    }
+
+if __name__ == "__main__":
+    uvicorn.run(app, host="0.0.0.0", port=8000)