add many tokens

app.py

@@ -4,43 +4,83 @@ import pandas as pd
 import numpy as np
 from datetime import datetime
 from flask import Flask, jsonify, Response
-from model import download_data, format_data, train_model
+from model import download_data, format_data, train_model, get_training_data_path
 from config import model_file_path
 
 app = Flask(__name__)
 
 
 def update_data():
-    """Download price data, format data and train model."""
+    """Download price data, format data and train model for each token."""
+    tokens = ["ETH", "BTC", "SOL", "BNB", "ARB"]
     download_data()
-    format_data()
-    train_model()
+    for token in tokens:
+        format_data(token)
+        train_model(token)
 
 
-def get_eth_inference():
-    """Load model and predict current price."""
-    with open(model_file_path, "rb") as f:
+def get_inference(token, period):
+    try:
+        model_path = model_file_path[token]
+
+        with open(model_path, "rb") as f:
             loaded_model = pickle.load(f)
 
-    now_timestamp = pd.Timestamp(datetime.now()).timestamp()
-    X_new = np.array([now_timestamp]).reshape(-1, 1)
-    current_price_pred = loaded_model.predict(X_new)
+        # Загружаем последние данные для данного токена
+        training_price_data_path = get_training_data_path(token)
+        price_data = pd.read_csv(training_price_data_path)
 
-    return current_price_pred[0]
+        # Используем последние значения признаков для предсказания
+        last_row = price_data.iloc[-1]
+        last_timestamp = last_row["timestamp"]
+
+        # Преобразуем период в секунды
+        period_seconds = convert_period_to_seconds(period)
+        new_timestamp = last_timestamp + period_seconds
+
+        # Формируем данные для предсказания с новым timestamp
+        X_new = np.array(
+            [
+                new_timestamp,
+                last_row["price_diff"],
+                last_row["volatility"],
+                last_row["volume"],
+                last_row["moving_avg_7"],
+                last_row["moving_avg_30"],
+            ]
+        ).reshape(1, -1)
+
+        # Делаем предсказание
+        future_price_pred = loaded_model.predict(X_new)
+        return future_price_pred[0]
+
+    except Exception as e:
+        print(f"Error during inference: {str(e)}")
+        raise
 
 
-@app.route("/inference/<string:token>")
-def generate_inference(token):
-    """Generate inference for given token."""
-    if not token or token != "ETH":
-        error_msg = "Token is required" if not token else "Token not supported"
-        return Response(json.dumps({"error": error_msg}), status=400, mimetype='application/json')
+def convert_period_to_seconds(period):
+    """Конвертируем период в секунды."""
+    if period.endswith("min"):
+        return int(period[:-3]) * 60
+    elif period.endswith("h"):
+        return int(period[:-1]) * 3600
+    elif period.endswith("d"):
+        return int(period[:-1]) * 86400
+    else:
+        raise ValueError(f"Unknown period format: {period}")
 
+
+@app.route("/inference/<string:token>/<string:period>")
+def generate_inference(token, period):
+    """Generate inference for given token and period."""
     try:
-        inference = get_eth_inference()
+        inference = get_inference(token, period)
         return Response(str(inference), status=200)
     except Exception as e:
-        return Response(json.dumps({"error": str(e)}), status=500, mimetype='application/json')
+        return Response(
+            json.dumps({"error": str(e)}), status=500, mimetype="application/json"
+        )
 
 
 @app.route("/update")
@@ -55,4 +95,4 @@ def update():
 
 if __name__ == "__main__":
     update_data()
-    app.run(host="0.0.0.0", port=8000)
+    app.run(host="0.0.0.0", port=8080)

config.py

@@ -2,4 +2,15 @@ import os
 
 app_base_path = os.getenv("APP_BASE_PATH", default=os.getcwd())
 data_base_path = os.path.join(app_base_path, "data")
-model_file_path = os.path.join(data_base_path, "model.pkl")
+
+model_file_path = {
+    "ETH": os.path.join(data_base_path, "eth_model.pkl"),
+    "BTC": os.path.join(data_base_path, "btc_model.pkl"),
+    "SOL": os.path.join(data_base_path, "sol_model.pkl"),
+    "BNB": os.path.join(data_base_path, "bnb_model.pkl"),
+    "ARB": os.path.join(data_base_path, "arb_model.pkl"),
+}
+
+
+def get_training_data_path(token):
+    return os.path.join(data_base_path, f"{token.lower()}_price_data.csv")

model.py

@@ -1,23 +1,28 @@
 import os
 import pickle
-from zipfile import ZipFile
-from datetime import datetime
-import pandas as pd
 import numpy as np
+from xgboost import XGBRegressor
+from zipfile import ZipFile
+from datetime import datetime, timedelta
+import pandas as pd
 from sklearn.model_selection import train_test_split
-from sklearn import linear_model
 from updater import download_binance_monthly_data, download_binance_daily_data
 from config import data_base_path, model_file_path
 
-
 binance_data_path = os.path.join(data_base_path, "binance/futures-klines")
-training_price_data_path = os.path.join(data_base_path, "eth_price_data.csv")
+
+
+def get_training_data_path(token):
+    """
+    Возвращает путь к файлу данных для указанного токена.
+    """
+    return os.path.join(data_base_path, f"{token}_price_data.csv")
 
 
 def download_data():
     cm_or_um = "um"
-    symbols = ["ETHUSDT"]
-    intervals = ["1d"]
+    symbols = ["ETHUSDT", "BTCUSDT", "SOLUSDT", "BNBUSDT", "ARBUSDT"]
+    intervals = ["10min", "1d"]
     years = ["2020", "2021", "2022", "2023", "2024"]
     months = ["01", "02", "03", "04", "05", "06", "07", "08", "09", "10", "11", "12"]
     download_path = binance_data_path
@@ -34,20 +39,17 @@ def download_data():
     print(f"Downloaded daily data to {download_path}.")
 
 
-def format_data():
-    files = sorted([x for x in os.listdir(binance_data_path)])
+def format_data(token):
+    files = sorted(
+        [x for x in os.listdir(binance_data_path) if x.endswith(".zip") and token in x]
+    )
 
-    # No files to process
     if len(files) == 0:
         return
 
     price_df = pd.DataFrame()
     for file in files:
         zip_file_path = os.path.join(binance_data_path, file)
-
-        if not zip_file_path.endswith(".zip"):
-            continue
-
         myzip = ZipFile(zip_file_path)
         with myzip.open(myzip.filelist[0]) as f:
             line = f.readline()
@@ -70,38 +72,53 @@ def format_data():
         df.index.name = "date"
         price_df = pd.concat([price_df, df])
 
+    price_df["timestamp"] = price_df.index.map(pd.Timestamp.timestamp)
+    price_df["price_diff"] = price_df["close"].diff()
+    price_df["volatility"] = (price_df["high"] - price_df["low"]) / price_df["open"]
+    price_df["volume"] = price_df["volume"]
+    price_df["moving_avg_7"] = price_df["close"].rolling(window=7).mean()
+    price_df["moving_avg_30"] = price_df["close"].rolling(window=30).mean()
+
+    # Удаляем строки с NaN значениями
+    price_df.dropna(inplace=True)
+
+    # Сохраняем данные
+    training_price_data_path = get_training_data_path(token)
     price_df.sort_index().to_csv(training_price_data_path)
 
 
-def train_model():
-    # Load the eth price data
+def train_model(token):
+    training_price_data_path = get_training_data_path(token)
     price_data = pd.read_csv(training_price_data_path)
-    df = pd.DataFrame()
 
-    # Convert 'date' to a numerical value (timestamp) we can use for regression
-    df["date"] = pd.to_datetime(price_data["date"])
-    df["date"] = df["date"].map(pd.Timestamp.timestamp)
+    # Используем дополнительные признаки
+    x = price_data[
+        [
+            "timestamp",
+            "price_diff",
+            "volatility",
+            "volume",
+            "moving_avg_7",
+            "moving_avg_30",
+        ]
+    ]
+    y = price_data["close"]
 
-    df["price"] = price_data[["open", "close", "high", "low"]].mean(axis=1)
+    x_train, x_test, y_train, y_test = train_test_split(
+        x, y, test_size=0.2, random_state=0
+    )
 
-    # Reshape the data to the shape expected by sklearn
-    x = df["date"].values.reshape(-1, 1)
-    y = df["price"].values.reshape(-1, 1)
-
-    # Split the data into training set and test set
-    x_train, _, y_train, _ = train_test_split(x, y, test_size=0.2, random_state=0)
-
-    # Train the model
-    print("Training model...")
-    model = linear_model.Lasso(alpha=0.1)
+    model = XGBRegressor()
     model.fit(x_train, y_train)
-    print("Model trained.")
 
-    # create the model's parent directory if it doesn't exist
-    os.makedirs(os.path.dirname(model_file_path), exist_ok=True)
+    token_model_path = model_file_path[token]
+    os.makedirs(os.path.dirname(token_model_path), exist_ok=True)
 
-    # Save the trained model to a file
-    with open(model_file_path, "wb") as f:
+    with open(token_model_path, "wb") as f:
         pickle.dump(model, f)
 
-    print(f"Trained model saved to {model_file_path}")
+    print(f"Trained model saved to {token_model_path}")
+
+    # Optional: Оценка модели
+    y_pred = model.predict(x_test)
+    print(f"Mean Absolute Error: {np.mean(np.abs(y_test - y_pred))}")

requirements.txt

@@ -5,3 +5,12 @@ pandas==2.1.3
 Requests==2.32.0
 scikit_learn==1.3.2
 werkzeug>=3.0.3 # not directly required, pinned by Snyk to avoid a vulnerability
+itsdangerous
+Jinja2
+MarkupSafe
+python-dateutil
+pytz
+scipy
+six
+scikit-learn
+xgboost