def cross_validate(self, dataset, num_folds):
self.rho_list = [0.1] * 12
self.alpha_list = [7.3476104736328125, 6.8925933837890625, 8.2190093994140625, 8.2137908935546875, 8.2814483642578125, 7.9523162841796875, 7.9453582763671875, 6.6873321533203125, 7.2606353759765625, 7.0821990966796875, 7.1634979248046875, 8.4375]
if 1:
dataset.createFolds(num_folds)
for rho in GlmNet.rho_values:
print "rho: %f" % (rho)
best_rmsle_list = [float("inf")] * 12
min_alpha_list = np.array([self.min_alpha] * 12)
max_alpha_list = np.array([self.max_alpha] * 12)
rho_list = [rho] * 12
self.search(dataset, num_folds, self.max_depth, self.max_width, min_alpha_list, max_alpha_list, rho_list)
def train(self, dataset):
if self.debug:
print "Training elastic net..."
self._train(dataset, self.alpha_list, self.rho_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.regressor_list[month_ind].predict(features)
return predictions
LearnerBase.register(GlmNet)
self._train(fold_train, k)
cur_score.addFold(fold_test.getSales(),
self.predict(fold_test))
for month_ind in range(12):
cur_rmsle = cur_score.getRMSLE(month_ind)
if cur_rmsle < best_rmsle_list[month_ind]:
best_rmsle_list[month_ind] = cur_rmsle
best_k_list[month_ind] = k
self.k_list = list(best_k_list)
if self.debug:
print "Best k-values by month: %s" % str(self.k_list)
def train(self, dataset):
self._train(dataset, None, k_list=self.k_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:,
month_ind] = self.knn_list[month_ind].predict(features)
return predictions
LearnerBase.register(NearestNeighbor)
def predict(self, x):
""" TODO fix this somehow. """
return 0.0
def grade(self, features, essay_set, domain, options={}):
"""Return an integer grade for each feature vector in the specified array."""
f = open('output/ds.set%d.dom%d.%s.matOut' % (essay_set, domain, options['postfix']), 'r')
scores = list()
for line in f.readlines():
scores.append(float(line))
if options['round']:
grades = [int(round(score)) for score in scores]
else:
grades = [self.curve.curve(score) for score in scores]
return np.asarray(grades)
def set_curve(self, scores, grade_counts):
"""Set curve with histogram."""
self.curve = Curve(scores, histogram=grade_counts)
LearnerBase.register(MatlabExample)
fold_train = dataset.getTrainFold(fold_ind)
fold_test = dataset.getTestFold(fold_ind)
self.rf = RandomForestClassifier(n_estimators=n, min_split=split)
self.rf.fit(fold_train.getFeatures(), fold_train.getLabels())
fold_score = Score.Score(fold_test.getLabels(), self.predict(fold_test.getFeatures()))
prediction_inds = dataset.getTestFoldInds(fold_ind)
learner_predictions[prediction_inds] = self.predict(fold_test.getFeatures())
cur_score = Score.Score(dataset.getLabels(), learner_predictions).getLogLoss()
if cur_score < best_score:
if self.debug:
print "Achieved new best score %f" % cur_score
best_params = (n, split)
best_score = cur_score
self.n_estimators, self.min_split = best_params
def train(self, features, labels):
if self.debug:
print "Training random forest with n_estimators=%d, min_split=%d" % (self.n_estimators, self.min_split)
self.rf = RandomForestClassifier(n_estimators=self.n_estimators, min_split=self.min_split)
self.rf.fit(features, labels)
def predict(self, features):
num_samples, num_features = features.shape
probs = [prob[1] for prob in self.rf.predict_proba(features)]
return np.minimum(np.maximum(probs, 0.01 * np.ones(num_samples)), 0.99 * np.ones(num_samples))
LearnerBase.register(RandomForest)
print "Learning on month %d of 12 with %d samples..." %(month_ind+1, num_samples)
A = np.hstack((month_features, np.ones((num_samples,1))))
month_params, residues, rank, s = linalg.lstsq(A, dataset.getSalesForMonth(month_ind))
params[:, month_ind] = month_params
self.params = params
def predict(self, dataset):
features = dataset.getQuantitativeFeatures()
num_samples, num_features = features.shape
A = np.hstack((features, np.ones((num_samples,1))))
sales = A.dot(self.params)
return np.maximum(sales, np.zeros(sales.shape))
LearnerBase.register(QuantLinearRegression)
if cur_score < best_score:
if self.debug:
print "Achieved new best score %f" %cur_score
best_k = k
best_score = cur_score
self.k = best_k
def train(self, features, labels):
self._train_with_k(features, labels, self.k)
def predict(self, features):
num_samples, num_features = features.shape
A = np.hstack((features, np.ones((num_samples,1))))
probs = A.dot(self.params)
return np.minimum(np.maximum(probs, 0.01*np.ones(num_samples)), 0.99*np.ones(num_samples))
LearnerBase.register(RidgeRegression)
if self.debug:
print "Train SuportVectorMachines with %d features..." % (
dataset.getNumFeatures())
self.svr_list = []
for month_ind in range(12):
month_features = dataset.getFeaturesForMonth(month_ind)
if self.debug:
num_samples = month_features.shape[0]
print "Learning on month %d of 12 with %d samples..." % (
month_ind + 1, num_samples)
svr = SVR()
svr.fit(month_features, dataset.getSalesForMonth(month_ind))
self.svr_list.append(svr)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:,
month_ind] = self.svr_list[month_ind].predict(features)
return predictions
LearnerBase.register(SupportVectorMachines)
self.best_rmsle_list = [float("inf")] * 12
self.min_n_estimators_list = np.array([self.min_n_estimators] * 12)
self.max_n_estimators_list = np.array([self.max_n_estimators] * 12)
self.search(
dataset,
num_folds,
self.max_depth,
self.max_width,
self.min_n_estimators_list,
self.max_n_estimators_list,
)
def train(self, dataset):
if self.debug:
print "Training GradientBoosting model with %d features..." % (dataset.getNumFeatures())
self._train(dataset, self.n_estimators_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.regressor_list[month_ind].predict(features)
return predictions
LearnerBase.register(GradientBoosting)
def cross_validate(self, dataset, num_folds):
pass
def train(self, dataset):
if self.debug:
print "Train SuportVectorMachines with %d features..." %(dataset.getNumFeatures())
self.svr_list = []
for month_ind in range(12):
month_features = dataset.getFeaturesForMonth(month_ind)
if self.debug:
num_samples = month_features.shape[0]
print "Learning on month %d of 12 with %d samples..." %(month_ind+1, num_samples)
svr = SVR()
svr.fit(month_features, dataset.getSalesForMonth(month_ind))
self.svr_list.append(svr)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.svr_list[month_ind].predict(features)
return predictions
LearnerBase.register(SupportVectorMachines)
if self.debug:
print "Running linear regression with %d quantitative features..." % (
num_features)
for month_ind in range(12):
month_features = dataset.getQuantitativeFeaturesForMonth(month_ind)
num_samples = month_features.shape[0]
if self.debug:
print "Learning on month %d of 12 with %d samples..." % (
month_ind + 1, num_samples)
A = np.hstack((month_features, np.ones((num_samples, 1))))
month_params, residues, rank, s = linalg.lstsq(
A, dataset.getSalesForMonth(month_ind))
params[:, month_ind] = month_params
self.params = params
def predict(self, dataset):
features = dataset.getQuantitativeFeatures()
num_samples, num_features = features.shape
A = np.hstack((features, np.ones((num_samples, 1))))
sales = A.dot(self.params)
return np.maximum(sales, np.zeros(sales.shape))
LearnerBase.register(QuantLinearRegression)
for split in RandomForest.split_values:
print "Split: %d" % (split)
self.min_n_estimators_list = np.array([0] * 12)
self.max_n_estimators_list = np.array([150] * 12)
self.search(dataset, num_folds, self.max_depth, self.max_width,
self.min_n_estimators_list,
self.max_n_estimators_list, [split] * 12)
def train(self, dataset):
if self.debug:
print "Training random forest with n_estimators_list: %s, min_samples_split_list: %s" % (
str(self.n_estimators_list), str(self.min_samples_split_list))
self._train(dataset, self.n_estimators_list,
self.min_samples_split_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.regressor_list[month_ind].predict(
features)
return predictions
LearnerBase.register(RandomForest)
print "Training SVM regression with C=%d, poly_degree=%d" %(self.C, self.poly_degree)
self._train_with_values(dataset, poly_degree=self.poly_degree, C=self.C)
def predict(self, dataset):
assert self.svm_list is not None
self._format_test_data(dataset)
num_samples = dataset.getNumSamples()
num_features = dataset.getNumFeatures()
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
# import pdb;pdb.set_trace()
predictions[:, month_ind] = svmlight.classify(self.svm_list[month_ind], self.formatted_data)
return predictions
LearnerBase.register(SVMRegression)
for fold_ind in range(num_folds):
fold_train = dataset.getTrainFold(fold_ind)
fold_test = dataset.getTestFold(fold_ind)
self._train(fold_train, k)
cur_score.addFold(fold_test.getSales(), self.predict(fold_test))
for month_ind in range(12):
cur_rmsle = cur_score.getRMSLE(month_ind)
if cur_rmsle < best_rmsle_list[month_ind]:
best_rmsle_list[month_ind] = cur_rmsle
best_k_list[month_ind] = k
self.k_list = list(best_k_list)
if self.debug:
print "Best k-values by month: %s" %str(self.k_list)
def train(self, dataset):
self._train(dataset, None, k_list=self.k_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.knn_list[month_ind].predict(features)
return predictions
LearnerBase.register(NearestNeighbor)
def cross_validate(self, dataset, num_folds):
pass
def train(self, dataset):
if self.debug:
print "Train SuportVectorMachines with %d features..." %(dataset.getNumFeatures())
self.regressor_list = []
for month_ind in range(12):
month_features = dataset.getFeaturesForMonth(month_ind)
if self.debug:
num_samples = month_features.shape[0]
print "Learning on month %d of 12 with %d samples..." %(month_ind+1, num_samples)
regressor = DecisionTreeRegressor()
regressor.fit(month_features, dataset.getSalesForMonth(month_ind))
self.regressor_list.append(regressor)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.regressor_list[month_ind].predict(features)
return predictions
LearnerBase.register(DecisionTree)
]
if 1:
dataset.createFolds(num_folds)
self.best_rmsle_list = [float("inf")] * 12
self.min_n_estimators_list = np.array([self.min_n_estimators] * 12)
self.max_n_estimators_list = np.array([self.max_n_estimators] * 12)
self.search(dataset, num_folds, self.max_depth, self.max_width,
self.min_n_estimators_list, self.max_n_estimators_list)
def train(self, dataset):
if self.debug:
print "Training GradientBoosting model with %d features..." % (
dataset.getNumFeatures())
self._train(dataset, self.n_estimators_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.regressor_list[month_ind].predict(
features)
return predictions
LearnerBase.register(GradientBoosting)
print "Learning on month %d of 12 with %d samples..." %(month_ind+1, num_samples)
A = np.hstack((month_features, np.ones((num_samples,1))))
month_params, residues, rank, s = linalg.lstsq(A, dataset.getSalesForMonth(month_ind))
params[:, month_ind] = month_params
self.params = params
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
A = np.hstack((features, np.ones((num_samples,1))))
sales = A.dot(self.params)
return np.maximum(sales, np.zeros(sales.shape))
LearnerBase.register(LinearRegression)
if 1:
dataset.createFolds(num_folds)
for rho in GlmNet.rho_values:
print "rho: %f" % (rho)
best_rmsle_list = [float("inf")] * 12
min_alpha_list = np.array([self.min_alpha] * 12)
max_alpha_list = np.array([self.max_alpha] * 12)
rho_list = [rho] * 12
self.search(dataset, num_folds, self.max_depth, self.max_width,
min_alpha_list, max_alpha_list, rho_list)
def train(self, dataset):
if self.debug:
print "Training elastic net..."
self._train(dataset, self.alpha_list, self.rho_list)
def predict(self, dataset):
features = dataset.getFeatures()
num_samples, num_features = features.shape
predictions = np.zeros((num_samples, 12))
for month_ind in range(12):
predictions[:, month_ind] = self.regressor_list[month_ind].predict(
features)
return predictions
LearnerBase.register(GlmNet)