advanced-algorithmic-trading

255
The same approach is carried out for Random Forests. However, since Random Forests
implicitly use a regression tree as their base estimators, there is no need to specify it in the
ensemble constructor:
# Estimate the Random Forest MSE over the full number
# of estimators, across a step size ("step_factor")
for i in range(0, axis_step):
print("Random Forest Estimator: %d of %d..." % (
step_factor*(i+1), n_estimators)
)
rf = RandomForestRegressor(
n_estimators=step_factor*(i+1),
n_jobs=n_jobs,
random_state=random_state
)
rf.fit(X_train, y_train)
mse = mean_squared_error(y_test, rf.predict(X_test))
estimators[i] = step_factor*(i+1)
rf_mse[i] = mse
Similarly for the AdaBoost boosting algorithm, with the exception that n_jobs is not present
due to the lack of parallelisability of boosting techniques. The learning rate, or shrinkage factor,
λ has been set to 0.01. Adjusting this value has a large impact on the absolute MSE calculated
for each estimator total:
# Estimate the AdaBoost MSE over the full number
# of estimators, across a step size ("step_factor")
for i in range(0, axis_step):
print("Boosting Estimator: %d of %d..." % (
step_factor*(i+1), n_estimators)
)
boosting = AdaBoostRegressor(
DecisionTreeRegressor(),
n_estimators=step_factor*(i+1),
random_state=random_state,
learning_rate=0.01
)
boosting.fit(X_train, y_train)
mse = mean_squared_error(y_test, boosting.predict(X_test))
estimators[i] = step_factor*(i+1)
boosting_mse[i] = mse
The final snippet of code simply plots these arrays against each other using Matplotlib but
with Seaborn’s default colour scheme. This is more visually pleasing than the Matplotlib defaults:
# Plot the chart of MSE versus number of estimators
plt.figure(figsize=(8, 8))
plt.title(’Bagging, Random Forest and Boosting comparison’)