Custom backtest for Time Series model validation

Custom_backtest_for_time_series.py

import numpy as np
import pandas as pd
import requests
from sklearn.linear_model import BayesianRidge

#load market data
market_data = get_market_data('BTC/USD')

#shift target variable one day back 
#in order to train model to predict one day into future
market_data['low_shifted'] = market_data['low'].shift(-1)
market_data = market_data.dropna()

#make numpy arrays
x = np.array(market_data[['close', 'high', 'low', 'open']])
y = np.array(market_data['low_shifted']).reshape(-1,1)

#custom backtest function
def backtest(n_days):
    """
    n_days - amount of the last n_days that we whant to get prediction and calculate metrics
    """
    predictions = []
    true_values = []
    for i in reversed(range(1, n_days)):
        x_train = x[:len(x)-i]
        y_train = y[:len(y)-i]
        x_test = x[len(x)-i]
        y_test = y[len(y)-i]
        model = BayesianRidge()
        model.fit(x_train, y_train)
        predictions.append(model.predict([x_test])[0])
        true_values.append(y_test[0])
    return true_values, predictions
 
true_values, predictions = backtest(20)