app

app.py

@@ -0,0 +1,138 @@
+################### PACKAGES
+import ccxt
+import pandas as pd
+import numpy as np
+from datetime import datetime, timedelta
+import time
+import requests
+from pytrends.request import TrendReq
+import joblib
+import gradio as gr
+from sklearn.preprocessing import StandardScaler
+
+
+################### MODEL LOADING
+# Load XGBoost and NGBoost
+xgb_artifact = joblib.load("xgb_volatility_model.joblib")
+xgb_model = xgb_artifact["model"]
+xgb_features = xgb_artifact["feature_names"]
+
+ngb_artifact = joblib.load("ngb_volatility_model.joblib")
+ngb_model = ngb_artifact["model"]
+ngb_features = ngb_artifact["feature_names"]
+
+# Load KMeans + scaler
+kmeans_artifact = joblib.load("kmeans_model.joblib")
+kmeans_model = kmeans_artifact["model"]
+cluster_scaler = kmeans_artifact["scaler"]
+cluster_features_names = kmeans_artifact["feature_names"]
+
+
+################### FEATURE GATHERING
+def fetch_ohlcv_last_6_days(symbol="BTC/USDT", date="2026-01-18", timeframe="1d"):
+    binance = ccxt.binance()
+    user_date = pd.to_datetime(date)
+    start_date = user_date - timedelta(days=5)  # 5 days before
+    since = binance.parse8601(start_date.strftime("%Y-%m-%dT00:00:00Z"))
+    data = binance.fetch_ohlcv(symbol, timeframe=timeframe, since=since, limit=6)
+    df = pd.DataFrame(data, columns=["timestamp","open","high","low","close","volume"])
+    df["timestamp"] = pd.to_datetime(df["timestamp"], unit="ms")
+    df.set_index("timestamp", inplace=True)
+    return df
+
+def fetch_fg_last_6_days(df):
+    url = "https://api.alternative.me/fng/?limit=0&format=json"
+    response = requests.get(url)
+    fg_df = pd.DataFrame(response.json()["data"])
+    fg_df["timestamp"] = pd.to_datetime(fg_df["timestamp"], unit="s")
+    fg_df.set_index("timestamp", inplace=True)
+    fg_df = fg_df[["value"]].astype(float)
+    fg_df.rename(columns={"value":"fg_index"}, inplace=True)
+    fg_df = fg_df.reindex(df.index, method="ffill")
+    df["fg_index"] = fg_df["fg_index"].values
+    return df
+
+def fetch_google_trend_for_date(keyword="Bitcoin", target_date="2024-01-15", window=7):
+    pytrends = TrendReq(hl="en-US", tz=360)
+    target_date = pd.to_datetime(target_date)
+    start_date = (target_date - timedelta(days=window)).strftime("%Y-%m-%d")
+    end_date = target_date.strftime("%Y-%m-%d")
+    timeframe = f"{start_date} {end_date}"
+    try:
+        pytrends.build_payload([keyword], timeframe=timeframe)
+        df = pytrends.interest_over_time()
+        if df.empty:
+            return 0
+        df = df.rename(columns={keyword: "trend"})
+        df = df.drop(columns=["isPartial"], errors="ignore")
+        df.index = pd.to_datetime(df.index)
+        if target_date in df.index:
+            return df.loc[target_date, "trend"]
+        else:
+            return 0
+    except:
+        return 0
+
+
+################### FEATURE ENGINEERING
+def engineer_features(df):
+    df["log_return"] = np.log(df["close"] / df["close"].shift(1))
+    df["hl_spread"] = df["high"] - df["low"]
+    df["co_spread"] = df["close"] - df["open"]
+    df["momentum_3"] = df["close"] - df["close"].shift(3)
+    df["vol_change"] = df["volume"] - df["volume"].shift(1)
+    df["rolling_std_5"] = df["log_return"].rolling(5).std()
+    # Fill NaNs for first few rows
+    df = df.fillna(0)
+    return df
+
+################### ASSIGN CLUSTER
+def assign_cluster(df):
+    cluster_features = df[cluster_features_names]
+    scaled = cluster_scaler.transform(cluster_features)
+    df["cluster"] = kmeans_model.predict(scaled)
+    return df
+
+
+################### PREDICTION FUNCTION
+def predict_volatility(date):
+    #Fetch raw data
+    df = fetch_ohlcv_last_6_days(date=date)
+    df = fetch_fg_last_6_days(df)
+    trend_value = fetch_google_trend_for_date("Bitcoin", date)
+    df["trend"] = trend_value
+    df["trend"] = df["trend"].fillna(method="ffill").fillna(0)
+    
+    #Feature engineering
+    df = engineer_features(df)
+    
+    #Clustering
+    df = assign_cluster(df)
+    
+    #Select features for models (last row only)
+    X_xgb = df.iloc[-1:][xgb_features]
+    X_ngb = df.iloc[-1:][ngb_features]
+    
+    #Predict
+    point = xgb_model.predict(X_xgb)[0]
+    dist = ngb_model.pred_dist(X_ngb)
+    low, high = dist.ppf([0.025, 0.975])[0]
+    
+    return point, low, high
+
+
+################### GRADIO INTERFACE
+def gradio_predict(date):
+    point, low, high = predict_volatility(date)
+    return f"Point volatility: {point:.4f}\n95% CI: [{low:.4f}, {high:.4f}]"
+
+demo = gr.Interface(
+    fn=gradio_predict,
+    inputs=gr.Textbox(label="Date (YYYY-MM-DD)"),
+    outputs=gr.Textbox(label="Prediction"),
+    title="Crypto Volatility Predictor",
+    description="Enter a date to get predicted volatility and 95% confidence interval."
+)
+
+if __name__ == "__main__":
+    demo.launch()