Here, we'll provide a brief explanation of how the simulator works. It is implemented as a function that accepts our standard Pandas DataFrame data and the model's predictions as inputs. First, we'll define the buying threshold and the stake size (how much money we invest in each buy), as well as placeholder variables. The variables will be used to store the true and predicted time series, as well as the profits of our model (balances). Furthermore, we define the buy_price variable, which stores the price at which we bought the Bitcoins. If the price is 0, we assume that we do not hold any Bitcoins. The buy_points and sell_points lists indicate the points in time when we bought or sold the Bitcoins and are used only for plotting. Furthermore, we store the starting index, which is equivalent to the sliding window's size as shown in the following example:

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd

from sklearn import metrics

def simulate(data, preds):
 # Constants and placeholders
 buy_threshold = 0.5
 stake = 100

true, pred, balances = [], [], []

buy_price = 0
 buy_points, sell_points = [], []
 balance = 0

start_index = len(data)-len(preds)-1

Next, for each point, we store the actual and predicted values. If the predicted value is greater than 0.5 and we do not hold any Bitcoins, we buy 100 USD worth of Bitcoins. If the predicted value is less than -0.5 and we have already bought Bitcoins, we sell them at the current close value. We add the current profit (or loss) to our balances, cast the true and predicted values as NumPy arrays, and produce the plots:

# Calculate predicted values
 for i in range(len(preds)):

last_close = data.Close[i+start_index-1]
 current_close = data.Close[i+start_index]

# Save predicted values and true values
 true.append(current_close)
 pred.append(last_close*(1+preds[i]/100))

 # Buy/Sell according to signal
 if preds[i] > buy_threshold and buy_price == 0:
 buy_price = true[-1]
 buy_points.append(i)

elif preds[i] < -buy_threshold and not buy_price == 0:
 profit = (current_close - buy_price) * stake/buy_price
 balance += profit
 buy_price = 0
 sell_points.append(i)

balances.append(balance)
 true = np.array(true)
 pred = np.array(pred)

# Create plots
 plt.figure()

plt.subplot(2, 1, 1)
 plt.plot(true, label='True')
 plt.plot(pred, label='pred')
 plt.scatter(buy_points, true[buy_points]+500, marker='v',
 c='blue', s=5, zorder=10)
 plt.scatter(sell_points, true[sell_points]-500, marker='^'
 , c='red', s=5, zorder=10)
 plt.title('Trades')

plt.subplot(2, 1, 2)
 plt.plot(balances)
 plt.title('Profit')
 print('MSE: %.2f'%metrics.mean_squared_error(true, pred))
 balance_df = pd.DataFrame(balances)

pct_returns = balance_df.diff()/stake
 pct_returns = pct_returns[pct_returns != 0].dropna()

 print('Sharpe: %.2f'%(np.mean(pct_returns)/np.std(pct_returns)))