Spaces:
Sleeping
Sleeping
File size: 11,597 Bytes
902fa1b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 |
# src/api.py - Enhanced API with better error handling for patient data
from fastapi import FastAPI, HTTPException, File, UploadFile
from pydantic import BaseModel, Field
import torch
import numpy as np
import joblib
from src.model import TabularVAE
from typing import List, Optional, Dict, Any
import os
import shutil
from fastapi.responses import JSONResponse, HTMLResponse
import json
app = FastAPI(title="Healthcare VAE API", version="1.0.0")
# Load model and scaler
try:
# Load feature names and determine input dimension
if os.path.exists("models/feature_names.pkl"):
feature_names = joblib.load("models/feature_names.pkl")
INPUT_DIM = len(feature_names)
print(f"Loaded {INPUT_DIM} features: {feature_names}")
else:
# Fallback to default features
feature_names = ["age", "gender", "diagnosis", "blood_type", "length_of_stay",
"age_group", "admission_season", "admission_day", "admission_month", "admission_year"]
INPUT_DIM = len(feature_names)
print(f"Using default {INPUT_DIM} features")
LATENT_DIM = 8
model = TabularVAE(input_dim=INPUT_DIM, latent_dim=LATENT_DIM, hidden_dims=(32, 16))
model.load_state_dict(torch.load("models/vae_model.pth", map_location='cpu'))
model.eval()
scaler = joblib.load("models/scaler.pkl")
# Load encoders if available
encoders = None
if os.path.exists("models/encoders.pkl"):
encoders = joblib.load("models/encoders.pkl")
print("Model and scaler loaded successfully!")
except Exception as e:
print(f"Error loading model: {e}")
print("Please run training first!")
class GenerateRequest(BaseModel):
n_samples: int = Field(..., ge=1, le=1000, description="Number of samples to generate")
random_seed: Optional[int] = Field(None, description="Random seed for reproducibility")
temperature: float = Field(1.0, ge=0.1, le=2.0, description="Sampling temperature")
class PatientData(BaseModel):
age: float = Field(..., ge=0, le=120, description="Patient age")
gender: str = Field(..., description="Patient gender (Male/Female)")
diagnosis: str = Field(..., description="Patient diagnosis")
blood_type: str = Field(..., description="Blood type")
length_of_stay: Optional[float] = Field(None, description="Length of stay in days")
age_group: Optional[int] = Field(None, ge=0, le=4, description="Age group (0-4)")
admission_season: Optional[int] = Field(None, ge=0, le=3, description="Admission season (0-3)")
admission_day: Optional[int] = Field(None, ge=0, le=6, description="Admission day of week (0-6)")
admission_month: Optional[int] = Field(None, ge=0, le=11, description="Admission month (0-11)")
admission_year: Optional[int] = Field(None, description="Admission year (normalized)")
class GeneratedResponse(BaseModel):
data: List[List[float]]
metadata: dict
def convert_numpy_to_python(obj):
"""Convert numpy types to Python native types for JSON serialization"""
if isinstance(obj, np.integer):
return int(obj)
elif isinstance(obj, np.floating):
return float(obj)
elif isinstance(obj, np.ndarray):
return obj.tolist()
elif isinstance(obj, list):
return [convert_numpy_to_python(item) for item in obj]
elif isinstance(obj, dict):
return {key: convert_numpy_to_python(value) for key, value in obj.items()}
else:
return obj
@app.get("/")
def read_root():
return {"message": "Healthcare VAE API is running!", "features": feature_names}
@app.get("/features")
def get_features():
"""Get information about the model features"""
return {
"feature_names": feature_names,
"input_dim": INPUT_DIM,
"latent_dim": LATENT_DIM
}
@app.post("/generate", response_model=GeneratedResponse)
def generate_synthetic_data(request: GenerateRequest):
try:
if request.random_seed is not None:
torch.manual_seed(request.random_seed)
np.random.seed(request.random_seed)
# Generate samples
z = torch.randn(request.n_samples, LATENT_DIM) * request.temperature
with torch.no_grad():
samples = model.decode(z).numpy()
# Inverse transform to original scale
data = scaler.inverse_transform(samples).tolist()
metadata = {
"n_samples": request.n_samples,
"latent_dim": LATENT_DIM,
"temperature": request.temperature,
"features": feature_names
}
return {"data": data, "metadata": metadata}
except Exception as e:
raise HTTPException(status_code=500, detail=f"Generation failed: {str(e)}")
@app.post("/encode")
def encode_patient(patient: PatientData):
"""Encode patient data to latent space"""
try:
# Convert patient data to feature vector
feature_vector = []
# Age
feature_vector.append(patient.age)
# Gender (encode if encoders available)
if encoders and 'gender' in encoders:
gender_encoded = encoders['gender'].transform([patient.gender])[0]
feature_vector.append(gender_encoded)
else:
# Fallback encoding
gender_encoded = 0 if patient.gender.lower() == 'male' else 1
feature_vector.append(gender_encoded)
# Diagnosis (encode if encoders available)
if encoders and 'diagnosis' in encoders:
diagnosis_encoded = encoders['diagnosis'].transform([patient.diagnosis])[0]
feature_vector.append(diagnosis_encoded)
else:
# Fallback encoding (simple hash)
diagnosis_encoded = hash(patient.diagnosis) % 10
feature_vector.append(diagnosis_encoded)
# Blood type (encode if encoders available)
if encoders and 'blood_type' in encoders:
blood_encoded = encoders['blood_type'].transform([patient.blood_type])[0]
feature_vector.append(blood_encoded)
else:
# Fallback encoding (simple hash)
blood_encoded = hash(patient.blood_type) % 8
feature_vector.append(blood_encoded)
# Length of stay
los = patient.length_of_stay if patient.length_of_stay is not None else 7.0
feature_vector.append(los)
# Age group
age_group = patient.age_group if patient.age_group is not None else 2
feature_vector.append(age_group)
# Admission season
season = patient.admission_season if patient.admission_season is not None else 0
feature_vector.append(season)
# Admission day
day = patient.admission_day if patient.admission_day is not None else 0
feature_vector.append(day)
# Admission month
month = patient.admission_month if patient.admission_month is not None else 0
feature_vector.append(month)
# Admission year
year = patient.admission_year if patient.admission_year is not None else 4
feature_vector.append(year)
# Ensure we have the right number of features
if len(feature_vector) != INPUT_DIM:
# Pad or truncate to match input dimension
while len(feature_vector) < INPUT_DIM:
feature_vector.append(0.0)
feature_vector = feature_vector[:INPUT_DIM]
# Convert to array and scale
data = np.array([feature_vector])
scaled_data = scaler.transform(data)
tensor_data = torch.tensor(scaled_data, dtype=torch.float32)
with torch.no_grad():
mu, logvar = model.encode(tensor_data)
# Convert numpy types to Python native types for JSON serialization
response = {
"latent_mean": convert_numpy_to_python(mu.numpy().tolist()),
"latent_logvar": convert_numpy_to_python(logvar.numpy().tolist()),
"features_used": feature_names,
"feature_values": convert_numpy_to_python(feature_vector)
}
return response
except Exception as e:
raise HTTPException(status_code=500, detail=f"Encoding failed: {str(e)}")
@app.get("/health")
def health_check():
"""Health check endpoint"""
return {
"status": "healthy",
"model_loaded": True,
"input_dim": INPUT_DIM,
"latent_dim": LATENT_DIM
}
@app.post("/upload_data")
async def upload_data(file: UploadFile = File(...)):
"""Upload a CSV file for continual training."""
os.makedirs("data", exist_ok=True)
file_location = "data/new_data.csv"
with open(file_location, "wb") as buffer:
shutil.copyfileobj(file.file, buffer)
return {"status": "success", "filename": file.filename}
@app.get("/training_progress")
def get_training_progress():
"""Get the latest training progress metrics for the web interface."""
progress_file = "data/training_progress.json"
if not os.path.exists(progress_file):
return JSONResponse(content={"status": "no_progress", "message": "No training progress found."}, status_code=404)
with open(progress_file, "r") as f:
progress = json.load(f)
return JSONResponse(content=progress)
@app.get("/dashboard", response_class=HTMLResponse)
def dashboard():
html = '''
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>Training Progress Dashboard</title>
<style>
body { font-family: Arial, sans-serif; margin: 2em; background: #f9f9f9; }
h1 { color: #2c3e50; }
#progress { background: #fff; padding: 1em; border-radius: 8px; box-shadow: 0 2px 8px #eee; max-width: 400px; }
.label { color: #888; }
</style>
</head>
<body>
<h1>Training Progress</h1>
<div id="progress">
<div><span class="label">Epoch:</span> <span id="epoch">-</span></div>
<div><span class="label">Train Loss:</span> <span id="train_loss">-</span></div>
<div><span class="label">Val Loss:</span> <span id="val_loss">-</span></div>
<div><span class="label">Best Val Loss:</span> <span id="best_val_loss">-</span></div>
<div><span class="label">Last Updated:</span> <span id="timestamp">-</span></div>
</div>
<script>
async function fetchProgress() {
try {
const res = await fetch('/training_progress');
if (!res.ok) throw new Error('No progress yet');
const data = await res.json();
document.getElementById('epoch').textContent = data.epoch;
document.getElementById('train_loss').textContent = data.train_loss?.toFixed(4);
document.getElementById('val_loss').textContent = data.val_loss?.toFixed(4);
document.getElementById('best_val_loss').textContent = data.best_val_loss?.toFixed(4);
const date = new Date(data.timestamp * 1000);
document.getElementById('timestamp').textContent = date.toLocaleString();
} catch (e) {
document.getElementById('progress').innerHTML = '<b>No training progress yet.</b>';
}
}
fetchProgress();
setInterval(fetchProgress, 3000);
</script>
</body>
</html>
'''
return HTMLResponse(content=html)
# Run with: uvicorn src.api:app --reload --host 0.0.0.0 --port 8000 |