Merge pull request #41 from 10-academy-w-9/time-series-forcasting

Time series forcasting using moirai model

requirements.txt

@@ -18,4 +18,17 @@ python-dotenv
 ccxt
 yfinance
 flask_cors
-Flask-Migrate
+safetensors
+torch>=2.1
+lightning>=2.0
+gluonts~=0.14.3
+# numpy~=1.26.0
+einops==0.7.*
+jaxtyping~=0.2.24
+python-dotenv
+hydra-core==1.3
+huggingface_hub>=0.23.0
+safetensors
+jax[cpu]
+Flask-Migrate
+

scripts/forecast_backtest_runner.py

@@ -0,0 +1,268 @@
+import os
+import sys 
+import matplotlib.pyplot as plt
+root_path = os.path.abspath(os.path.join(os.getcwd(), '../notebooks/uni2ts/src'))
+sys.path.append(root_path)
+
+# root_path = os.path.abspath(os.path.join(os.getcwd(), './uni2ts/src'))
+# sys.path.append(root_path)
+import matplotlib.pyplot as plt
+import pandas as pd
+import numpy as np
+import torch
+from einops import rearrange
+from gluonts.dataset.pandas import PandasDataset
+from gluonts.dataset.split import split
+from gluonts.torch.model.predictor import PyTorchPredictor
+import yfinance as yf
+
+from uni2ts.eval_util.plot import plot_single
+from uni2ts.model.moirai import MoiraiForecast, MoiraiModule
+import backtrader as bt
+
+# Step 1: Fetch Data from Yahoo Finance
+def fetch_data(symbol, start_date, end_date):
+    try:
+        ticker = yf.Ticker(symbol)
+        ohlcv = ticker.history(start=start_date, end=end_date)
+        ohlcv.reset_index(inplace=True)
+        ohlcv['timestamp'] = pd.to_datetime(ohlcv['Date'])
+        ohlcv.set_index('timestamp', inplace=True)
+        ohlcv = ohlcv[['Open', 'High', 'Low', 'Close', 'Volume']]
+        ohlcv.rename(columns={'Open': 'open', 'High': 'high', 'Low': 'low', 'Close': 'close', 'Volume': 'volume'}, inplace=True)
+
+        # Ensure the data is uniformly spaced by resampling
+        ohlcv = ohlcv.resample('D').ffill()
+
+        return ohlcv
+    except Exception as e:
+        print(f"Error fetching data for {symbol}: {str(e)}")
+        return None
+
+def load_and_predict(data):
+    # Use only the 'close' price for forecasting
+    df = data[['close']].rename(columns={'close': 'target'})
+
+    # Ensure data is sorted
+    df = df.sort_index()
+
+    # Step 2: Prepare the Data for the Model
+    # Convert into GluonTS dataset
+    ds = PandasDataset(dict(df), freq="D")
+
+    # Split into train/test set
+    TEST = 100  # Define the length of the test set
+    PDT = 20  # Define the prediction length
+    train, test_template = split(ds, offset=-TEST)
+
+    # Ensure the length of the dataset is sufficient for rolling window evaluation
+    total_length = len(df)
+    if total_length < TEST + PDT:
+        raise ValueError(f"Not enough data points. Total length: {total_length}, TEST: {TEST}, PDT: {PDT}")
+
+    # Construct rolling window evaluation
+    test_data = test_template.generate_instances(
+        prediction_length=PDT,
+        windows=TEST // PDT,
+        distance=PDT
+    )
+
+    # Step 3: Load the Model and Make Predictions
+    SIZE = "small"  # Model size
+    CTX = 200  # Context length
+    PSZ = "auto"  # Patch size
+    BSZ = 32  # Batch size
+
+    # Prepare pre-trained model by downloading model weights from Hugging Face hub
+    model = MoiraiForecast(
+        module=MoiraiModule.from_pretrained(f"Salesforce/moirai-1.0-R-{SIZE}"),
+        prediction_length=PDT,
+        context_length=CTX,
+        patch_size=PSZ,
+        num_samples=100,
+        target_dim=1,
+        feat_dynamic_real_dim=ds.num_feat_dynamic_real,
+        past_feat_dynamic_real_dim=ds.num_past_feat_dynamic_real,
+    )
+
+    predictor = model.create_predictor(batch_size=BSZ)
+    forecasts = predictor.predict(test_data.input)
+
+    # Step 4: Plot the Results
+    input_it = iter(test_data.input)
+    label_it = iter(test_data.label)
+    forecast_it = iter(forecasts)
+
+    try:
+        inp = next(input_it)
+        label = next(label_it)
+        forecast = next(forecast_it)
+
+        plot_single(
+            inp, 
+            label, 
+            forecast, 
+            context_length=CTX,
+            name="pred",
+            show_label=True,
+        )
+        plt.show()
+    except StopIteration:
+        print("Error: Not enough data points to generate forecasts.")
+
+    return forecasts
+
+# Define the strategies
+class RsiBollingerBandsStrategy(bt.Strategy):
+    params = (
+        ('rsi_period', 14),
+        ('bb_period', 20),
+        ('bb_dev', 2),
+        ('oversold', 30),
+        ('overbought', 70),
+    )
+
+    def __init__(self, predictions):
+        self.rsi = bt.indicators.RelativeStrengthIndex(period=self.params.rsi_period)
+        self.bbands = bt.indicators.BollingerBands(period=self.params.bb_period, devfactor=self.params.bb_dev)
+        self.predictions = predictions
+        self.current_prediction = next(iter(self.predictions))
+
+    def next(self):
+        if not self.position:
+            if self.rsi < self.params.oversold and self.data.close <= self.bbands.lines.bot:
+                self.buy()
+        else:
+            if self.rsi > self.params.overbought or self.data.close >= self.bbands.lines.top:
+                self.sell()
+
+        # Move to the next prediction if available
+        try:
+            self.current_prediction = next(iter(self.predictions))
+        except StopIteration:
+            pass
+
+class MacdStrategy(bt.Strategy):
+    params = (
+        ('macd1_period', 12),
+        ('macd2_period', 26),
+        ('signal_period', 9),
+    )
+
+    def __init__(self, predictions):
+        self.macd = bt.indicators.MACDHisto(period_me1=self.params.macd1_period, period_me2=self.params.macd2_period, period_signal=self.params.signal_period)
+        self.predictions = predictions
+        self.current_prediction = next(iter(self.predictions))
+
+    def next(self):
+        if not self.position:
+            if self.macd.lines.histo[0] > 0 and self.macd.lines.histo[-1] <= 0:
+                self.buy()
+        else:
+            if self.macd.lines.histo[0] < 0 and self.macd.lines.histo[-1] >= 0:
+                self.sell()
+
+        # Move to the next prediction if available
+        try:
+            self.current_prediction = next(iter(self.predictions))
+        except StopIteration:
+            pass
+
+class StochasticOscillatorStrategy(bt.Strategy):
+    params = (
+        ('stoch_period', 14),
+        ('stoch_low', 20),
+        ('stoch_high', 80),
+    )
+
+    def __init__(self, predictions):
+        self.stoch = bt.indicators.Stochastic(period=self.params.stoch_period)
+        self.predictions = predictions
+        self.current_prediction = next(iter(self.predictions))
+
+    def next(self):
+        if not self.position:
+            if self.stoch.lines.percK[0] < self.params.stoch_low and self.stoch.lines.percK[-1] >= self.params.stoch_low:
+                self.buy()
+        else:
+            if self.stoch.lines.percK[0] > self.params.stoch_high and self.stoch.lines.percK[-1] <= self.params.stoch_high:
+                self.sell()
+
+        # Move to the next prediction if available
+        try:
+            self.current_prediction = next(iter(self.predictions))
+        except StopIteration:
+            pass
+
+def run_backtest(strategy_class, symbol, start_date, end_date):
+    # Fetch data for backtesting
+    data = fetch_data(symbol, start_date, end_date)
+
+    # Load and make predictions
+    predictions = load_and_predict(data)
+
+    # Initialize cerebro
+    cerebro = bt.Cerebro()
+
+    # Add data feed
+    cerebro.adddata(bt.feeds.PandasData(dataname=data))
+
+    # Add strategy with predictions
+    cerebro.addstrategy(strategy_class, predictions=predictions)
+
+    # Set broker settings
+    cerebro.broker.set_cash(100000)
+    cerebro.broker.setcommission(commission=0.002)
+
+    # Add analyzers for performance metrics
+    cerebro.addanalyzer(bt.analyzers.SharpeRatio, _name='sharpe')
+    cerebro.addanalyzer(bt.analyzers.DrawDown, _name='drawdown')
+    cerebro.addanalyzer(bt.analyzers.TradeAnalyzer, _name='trades')
+    cerebro.addanalyzer(bt.analyzers.Returns, _name='returns')
+
+    # Print starting conditions
+    start_value = cerebro.broker.getvalue()
+    print(f'Starting Portfolio Value: {start_value:.2f}')
+
+    # Run backtest
+    results = cerebro.run()
+
+    # Print ending conditions
+    end_value = cerebro.broker.getvalue()
+    print(f'Ending Portfolio Value: {end_value:.2f}')
+
+    # Extracting backtest metrics
+    strat = results[0]
+
+    # Prepare results
+    result_dict = {
+        "Starting Portfolio Value": start_value,
+        "Ending Portfolio Value": end_value,
+        "Sharpe Ratio": strat.analyzers.sharpe.get_analysis().get('sharperatio', 'N/A'),
+        "Max Drawdown": strat.analyzers.drawdown.get_analysis().get('max', {}).get('drawdown', 'N/A'),
+        "Total Trades": strat.analyzers.trades.get_analysis().get('total', {}).get('total', 'N/A'),
+        "Winning Trades": strat.analyzers.trades.get_analysis().get('won', {}).get('total', 'N/A'),
+        "Losing Trades": strat.analyzers.trades.get_analysis().get('lost', {}).get('total', 'N/A'),
+        "Total Return": strat.analyzers.returns.get_analysis().get('rtot', 'N/A')
+    }
+
+    # Plot the results
+    cerebro.plot(style='candlestick')
+
+    # Print metrics
+    print("Metrics:")
+    for key, value in result_dict.items():
+        print(f"{key}: {value}")
+
+    return result_dict
+
+if __name__ == "__main__":
+    symbol = 'ETH-USD'
+    start_date = '2023-06-20'
+    end_date = '2024-06-20'
+
+    strategies = [RsiBollingerBandsStrategy, MacdStrategy, StochasticOscillatorStrategy]
+
+    for strategy in strategies:
+        print(f"Running backtest for {strategy.__name__}")
+        run_backtest(strategy, symbol, start_date, end_date)