def test_mape_with_perfect_prediction(self):
gaps = [1, 2, 3]
predictions = [1, 2, 3]
self.assertEqual(calculate_mape(gaps, predictions), 0)
def test_mape_with_zero_gaps(self):
gaps = [1, 0, 4]
predictions = [1, 2, 1]
self.assertEqual(calculate_mape(gaps, predictions), 1 / 3784.)
def test_mape_with_another_one_off(self):
gaps = [1, 2, 4]
predictions = [1, 2, 1]
self.assertEqual(calculate_mape(gaps, predictions), 1 / 3784.)
def test_mape_with_one_off(self):
gaps = [1, 2, 3]
predictions = [1, 2, 4]
self.assertEqual(calculate_mape(gaps, predictions), 1 / 8514.)
from sklearn import svm
from data_preprocessing.get_x_y_from_features import prediction_matching_data_split
from utilities.measures import calculate_mape
def train_predict():
svr = svm.SVR()
svr.fit(x_train, y_train)
return svr.predict(x_test)
# default SVR implementation
x_train, x_test, y_train, y_test = prediction_matching_data_split()
predictions = train_predict()
print calculate_mape(y_test, predictions)
training_error[i] = metrics.mean_absolute_error(
y_train[:s], dtr.predict(x_train[:s]))
test_error[i] = metrics.mean_absolute_error(
y_test, dtr.predict(x_test))
learning_curve_graph(sizes, training_error, test_error)
# use grid search to find the best depth
def find_best_depth(max_depths):
scorer = metrics.make_scorer(calculate_mape, greater_is_better=False)
model = grid_search.GridSearchCV(tree.DecisionTreeRegressor(),
{'max_depth': max_depths},
scorer,
n_jobs=-1,
cv=30)
model.fit(x_train, y_train)
return {
'prediction': model.predict(x_test),
'best_depth': model.best_params_['max_depth']
}
# split data into 75/25 percent sizes
x_train, x_test, y_train, y_test = prediction_matching_data_split()
# use grid search to get prediction results
gs_result = find_best_depth([d for d in range(10, 30)])
print 'best depth', gs_result['best_depth']
predictions = gs_result['prediction']
# use MAPE calculator to find the model performance
print 'MAPE: ', calculate_mape(y_test, predictions)