def eval(model, data, set_name, denorm_predictions=True):
# predictions
predictions = model.predict(data.dataset(set_name))
labels = data.raw_data(set_name)["labels"][:len(predictions)]
predictions = pd.DataFrame(data=predictions,
index=labels.index,
columns=labels.columns)
if denorm_predictions:
predictions = data.denormalize_labels(predictions)
# results
results = {
"general": {
"mae": mae(labels, predictions),
"mape": mape(labels, predictions),
"mse": mse(labels, predictions)
}
}
for col in labels.columns:
results[col] = {
"mae": mae(labels[col], predictions[col]),
"mape": mape(labels[col], predictions[col]),
"mse": mse(labels[col], predictions[col]),
"tend_acc": tendency_accuracy(labels[col], predictions[col])
}
return predictions, results
def linear_model(data):
"""Build, train, and test sklearn linear model."""
print('\nBuilding sklearn linear model with only numeric data ...')
print('#' * 65)
m = LinearRegression()
m.fit(data['x_train_numeric'], data['y_train'])
print('r^2 score:', m.score(data['x_test_numeric'], data['y_test']))
y_pred = m.predict(data['x_test_numeric'])
print('MSE:', mse(data['y_test'], y_pred))
print('MAE:', mae(data['y_test'], y_pred))
print('MAPE:', 100 * mape(data['y_test'], y_pred))
def evaluate(ticker):
'''Evaluate model performance'''
print('Retrieving data')
predictions = pd.concat([pd.read_csv(f, parse_dates=['ds']) for f in os.listdir() if f.endswith('.csv')])
df = yf.Ticker(ticker)\
.history(start=predictions.ds.min().to_pydatetime(),
end=predictions.ds.max().to_pydatetime(),
interval='1m')
df.index = df.index.floor('1min')
df.index = df.index.tz_convert('UTC')
df = df.loc[predictions.ds.min():predictions.ds.max()]
df = df.reset_index()
df = df.merge(predictions, left_on='Datetime', right_on='ds', how='inner')\
.rename(columns={'Close': 'Actuals', 'yhat': 'Predicted'})\
.set_index('Datetime')
df = df[['Actuals', 'Predicted']]
print('Metric calculation')
mape_value = mape(df.Actuals, df.Predicted) * 100
print('Plotting')
df.plot(title=f'Actuals vs. Predicted MAPE={round(mape_value, 2)}%')
plt.savefig('forecast_plot.png', dpi=400)
def lgb_mape(pred, real):
'''sklearn.metrics.mean_absolute_percentage_error wrapper for LGB'''
is_higher_better = False
score = mape(real.label, pred)
return 'lgb_mape', score, is_higher_better