app.py

import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import pandas as pd
from fastapi import FastAPI
from pydantic import BaseModel
from typing import List

# ==========================================
# 1. NETWORK DEFINITION
# ==========================================
class Actor(nn.Module):
    def __init__(self, state_dim, action_dim, hidden_dim=256):
        super().__init__()
        self.fc1 = nn.Linear(state_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, hidden_dim)
        self.fc3 = nn.Linear(hidden_dim, hidden_dim)
        self.mean = nn.Linear(hidden_dim, action_dim)
        self.log_std = nn.Linear(hidden_dim, action_dim)
        self.LOG_STD_MIN = -20
        self.LOG_STD_MAX = 2

    def forward(self, state):
        x = F.relu(self.fc1(state))
        x = F.relu(self.fc2(x))
        x = F.relu(self.fc3(x))
        mean = self.mean(x)
        log_std = self.log_std(x)
        log_std = torch.clamp(log_std, self.LOG_STD_MIN, self.LOG_STD_MAX)
        return mean, log_std

# ==========================================
# 2. DYNAMIC SCALER
# ==========================================
class DynamicScaler:
    def fit_transform(self, data):
        X = np.array(data, dtype=np.float32)
        mean = np.mean(X, axis=0)
        scale = np.std(X, axis=0)
        scale[scale < 1e-8] = 1.0
        X_scaled = (X - mean) / scale
        return np.clip(X_scaled, -5, 5)

# ==========================================
# 3. SETUP
# ==========================================
app = FastAPI()

# !!! THE FIX: 53 DIMENSIONS !!!
STATE_DIM = 53
ACTION_DIM = 1
HIDDEN_DIM = 256

device = torch.device("cpu")
actor = Actor(STATE_DIM, ACTION_DIM, HIDDEN_DIM).to(device)
scaler = DynamicScaler()

# Load Model
try:
    print("Loading model...")
    checkpoint = torch.load("sac_v9_pytorch_best_eval.pt", map_location=device)
    actor.load_state_dict(checkpoint['actor'])
    actor.eval()
    print(f"✅ Model loaded! Expecting {STATE_DIM} inputs.")
except Exception as e:
    print(f"❌ Error loading model: {e}")

# ==========================================
# 4. ENDPOINT
# ==========================================
class InputData(BaseModel):
    history: List[List[float]] 
    portfolio: List[float]

@app.post("/predict")
def predict(data: InputData):
    try:
        # 1. Scale History
        if len(data.history) < 2:
            return {"error": "Need at least 2 rows of history"}
            
        scaled_history = scaler.fit_transform(data.history)
        current_market = scaled_history[-1] # This likely has ~96 features
        
        # 2. Prepare Portfolio (5 items)
        portfolio = np.array(data.portfolio, dtype=np.float32)
        
        # 3. Smart Concatenation to match 53 Inputs
        # The model needs 53 inputs. 
        # Structure: [Market Features (48)] + [Portfolio (5)]
        
        needed_market = STATE_DIM - len(portfolio) # 48
        
        # Take the first 48 market features (most important ones like RSI, MACD usually first)
        if len(current_market) >= needed_market:
            market_part = current_market[:needed_market]
        else:
            # Pad with zeros if we somehow have less
            padding = np.zeros(needed_market - len(current_market))
            market_part = np.concatenate([current_market, padding])
            
        state = np.concatenate([market_part, portfolio])
        
        # 4. Inference
        state_tensor = torch.FloatTensor(state).unsqueeze(0).to(device)
        with torch.no_grad():
            mean, _ = actor(state_tensor)
            action = torch.tanh(mean).item()

        return {
            "action": action,
            "signal": "BUY" if action > 0 else "SELL",
            "confidence": abs(action)
        }

    except Exception as e:
        return {"error": str(e)}