CROSS_VALIDATION_FOLDS = 4
num_feature_list = [4,8,15]
depth_list = range(8,21,2) #range(1,8) + range(8,21,2)
accuracy_results = {}
for response in response_columns:
for num_features in num_feature_list:
feature_columns = all_feature_columns[:num_features]
for max_depth in depth_list:
print "---working on response %s | num_features %s | max_depth %s" % (response, num_features, max_depth)
clf = DecisionTreeClassifier(max_depth = max_depth)
X = one_minute_dataframe[feature_columns]
y = one_minute_dataframe[response]
cross_val_score_list = cross_val_score(clf, X, y, cv=tscvsplit(one_minute_dataframe,CROSS_VALIDATION_FOLDS), n_jobs = -1)
if response not in accuracy_results:
accuracy_results[response] = {num_features: {max_depth: np.mean(cross_val_score_list)}}
else:
if num_features not in accuracy_results[response]:
accuracy_results[response][num_features] = {max_depth: np.mean(cross_val_score_list)}
else:
accuracy_results[response][num_features][max_depth] = np.mean(cross_val_score_list)
print "response\tnum_features\tmax_depth\taccuracy"
for response in response_columns:
for num_features in num_feature_list:
for max_depth in depth_list:
print "%s\t%s\t\t%s\t\t%0.15f" % (response, num_features, max_depth, accuracy_results[response][num_features][max_depth])
z_results = {}
for response in response_columns:
for num_features in num_feature_list:
feature_columns = all_feature_columns[:num_features]
for max_depth in depth_list:
print "---working on response %s | num_features %s | max_depth %s" % (response, num_features, max_depth)
regressor = DecisionTreeRegressor(max_depth = max_depth)
X = one_minute_dataframe[feature_columns]
y = one_minute_dataframe[response]
# regressor.fit(X,y)
# trade_z_score_func(y, regressor.predict(X))
cross_val_z_list = cross_val_custom_score(regressor, X, y, tscvsplit(one_minute_dataframe,CROSS_VALIDATION_FOLDS), trade_z_score_func)
# cross_val_sse_list = cross_val_score(regressor, X, y, scoring=trade_z_score) #cv=tscvsplit(one_minute_dataframe,CROSS_VALIDATION_FOLDS), n_jobs = 1)
if response not in z_results:
z_results[response] = {num_features: {max_depth: np.mean(cross_val_z_list)}}
else:
if num_features not in z_results[response]:
z_results[response][num_features] = {max_depth: np.mean(cross_val_z_list)}
else:
z_results[response][num_features][max_depth] = np.mean(cross_val_z_list)
print "response\tnum_features\tmax_depth\ttrade_z"
for response in response_columns:
for num_features in num_feature_list:
for max_depth in depth_list:
print "%s\t%s\t\t%s\t\t%0.15f" % (response, num_features, max_depth, z_results[response][num_features][max_depth])
rmse_results = {}
for response in response_columns:
for num_features in num_feature_list:
feature_columns = all_feature_columns[:num_features]
for max_depth in depth_list:
print "---working on response %s | num_features %s | max_depth %s" % (response, num_features, max_depth)
regressor = DecisionTreeRegressor(max_depth=max_depth)
X = one_minute_dataframe[feature_columns]
y = one_minute_dataframe[response]
cross_val_sse_list = -cross_val_score(
regressor,
X,
y,
scoring="mean_squared_error",
cv=tscvsplit(one_minute_dataframe, CROSS_VALIDATION_FOLDS),
n_jobs=-1,
)
if response not in rmse_results:
rmse_results[response] = {
num_features: {max_depth: np.mean([np.sqrt(x) for x in cross_val_sse_list])}
}
else:
if num_features not in rmse_results[response]:
rmse_results[response][num_features] = {
max_depth: np.mean([np.sqrt(x) for x in cross_val_sse_list])
}
else:
rmse_results[response][num_features][max_depth] = np.mean(
[np.sqrt(x) for x in cross_val_sse_list]
)
