Random forests

Machine Learning para finanzas con Python

Nathan George

Data Science Professor

polynomial fit

linear fit

Random forests

random forest graphviz

Bootstrap aggregating (bagging)

bootstrapping

Feature sampling

Random Forests

A collection (ensemble) of decision trees
Bootstrap aggregating (bagging)
Sample of features at each split

sklearn implementation

from sklearn.ensemble import RandomForestRegressor

random_forest = RandomForestRegressor()
random_forest.fit(train_features, train_targets)
print(random_forest.score(train_features, train_targets))

Hyperparameters

random_forest = RandomForestRegressor(n_estimators=200,
                                      max_depth=5,
                                      max_features=4,
                                      random_state=42)

ParameterGrid

from sklearn.model_selection import ParameterGrid

grid = {'n_estimators': [200], 
        'max_depth':[3, 5], 
        'max_features': [4, 8]}


from pprint import pprint
pprint(list(ParameterGrid(grid)))

[{'max_depth': 3, 'max_features': 4, 'n_estimators': 200},
 {'max_depth': 3, 'max_features': 8, 'n_estimators': 200},
 {'max_depth': 5, 'max_features': 4, 'n_estimators': 200},
 {'max_depth': 5, 'max_features': 8, 'n_estimators': 200}]

ParameterGrid

test_scores = []
# loop through the parameter grid, set hyperparameters, save the scores
for g in ParameterGrid(grid):
    rfr.set_params(**g)  # ** is "unpacking" the dictionary
    rfr.fit(train_features, train_targets)
    test_scores.append(rfr.score(test_features, test_targets))

# find best hyperparameters from the test score and print
best_idx = np.argmax(test_scores)
print(test_scores[best_idx])
print(ParameterGrid(grid)[best_idx])

0.05594252725411142
{'max_depth': 5, 'max_features': 8, 'n_estimators': 200}

Plant some random forests!

Machine Learning para finanzas con Python