def run():
prep.preprocess_data()
X_train = prep.data_store.get_X_train(include_valid=True)
y_train = prep.data_store.get_y_train(include_valid=True,
one_hot_encoding=False)
X_test = prep.data_store.get_X_test()
y_test = prep.data_store.get_y_test(one_hot_encoding=False)
log.debug('X_train: %s, y_train: %s, X_test: %s, y_test: %s' %
(X_train.shape, y_train.shape, X_test.shape, y_test.shape))
log.debug('X_train[:5]: %s' % (X_train[100:105]))
log.debug('y_train[:5]: %s' % (y_train[100:105]))
pipeline = SVC()
train_sizes, train_scores, test_scores = \
learning_curve(estimator=pipeline,
X=X_train[:1000],
y=y_train[:1000],
train_sizes=np.linspace(0.1, 1.0, 10),
cv=10,
n_jobs=1)
train_mean = np.mean(train_scores, axis=1)
train_std = np.std(test_scores, axis=1)
test_mean = np.mean(test_scores, axis=1)
test_std = np.std(test_scores, axis=1)
log.debug('accuracy: %s, type: %s' % (train_mean[-1], type(train_mean)))
log.debug('accuracy: %s, type: %s' % (test_mean[-1], type(test_mean)))
log.debug("end")
def show_dimen_reduction(self, fig, ax):
log.debug('#####')
pk.preprocess_data()
self.get_dataset()
log.debug('X_train: %s, y_train: %s' %
(self.X_train.shape, self.y_train.shape))
clf, pipeline = self.build_clf_pipeline()
# rfecv = RFECV(estimator=clf, step=1, cv=5, scoring='accuracy')
# rfecv.fit(self.X_train, self.y_train)
# log.debug('Optimal number of features: %d' % rfecv.n_features_)
# Feature selector
selector_k_best = SelectKBest(f_regression, k=10)
# Build the machine learning pipeline
pipeline_clf = Pipeline([('selector', selector_k_best), ('rf', clf)])
best_features = None
best_score = 0
for k in range(2, pk.data_store.get_x_field_count() + 1):
pipeline_clf.set_params(selector__k=k)
# Training the classifier
pipeline_clf.fit(self.X_train, self.y_train)
# Print score
cur_score = pipeline_clf.score(self.X_test, self.y_test)
log.debug('Feature Count: %d, Score:%f' % (k, cur_score))
# Print the selected features chosen by the selector
features_status = pipeline_clf.named_steps['selector'].get_support(
)
selected_features = []
for count, item in enumerate(features_status):
if item:
selected_features.append(count)
log.debug('Selected features:%s' %
(','.join([str(x) for x in selected_features])))
if cur_score > best_score:
best_features = selected_features
best_score = cur_score
mdata.set_reduced_features(best_features)
def __init__(self, main_view=None, parent=None):
super(PreprocessSettingView, self).__init__(parent)
self.central_view = main_view
self.initialized = False
widget = QWidget()
layout = QVBoxLayout()
widget.setLayout(layout)
self.setWidget(widget)
self.combos = []
prep.preprocess_data()
self.X, self.y = prep.data_store.get_original_data()
self.central_view.set_data(self.X, self.y)
self.columns = self.X.columns.tolist()
for idx in range(len(self.columns)):
combo = QComboBox()
combo.addItems(self.columns + ['None'])
combo.setCurrentIndex(len(self.columns))
combo.currentTextChanged.connect(
partial(self.on_combo_changed, idx))
layout.addWidget(combo)
self.combos.append(combo)
def predict(self):
pk.preprocess_data()
self.get_dataset()
X_train, y_train, X_test, y_test = self.reduce_dataset()
log.debug('X_train: %s, y_train: %s, X_test: %s, y_test: %s' %
(X_train.shape, y_train.shape, X_test.shape, y_test.shape))
clf, pipeline = self.build_clf_pipeline()
self.set_best_params(clf, pipeline)
# Estimating model performance
pipeline.fit(X_train, y_train)
predicted = pipeline.predict(X_test)
score = AlgBase.accuracy(predicted, y_test)
log.debug('Test accuracy: %.2f' % (score))
sel_features = mdata.get_reduced_features()
log.debug('Selected features indeces: %s' % sel_features)
X_unseen = pk.data_store.get_X_unseen()
if sel_features is not None:
X_unseen = X_unseen[:, sel_features]
predictions = pipeline.predict(X_unseen).astype(int)
pk.data_store.save_predictions(predictions)
log.debug("end")
def show_learning_curve(self, fig, ax, best_params):
prep.preprocess_data()
batch_size = prep.data_store.get_batch_size()
num_features = prep.data_store.get_x_field_count()
num_labels = prep.data_store.get_y_value_count()
hidden_size = prep.data_store.get_hidden_size()
log.debug('num_features: %d, num_labels: %d' %
(num_features, num_labels))
train_dataset = prep.data_store.get_X_train(include_valid=False)
train_labels = prep.data_store.get_y_train(include_valid=False,
one_hot_encoding=True)
valid_dataset = prep.data_store.get_X_valid()
valid_labels = prep.data_store.get_y_valid(True)
test_dataset = prep.data_store.get_X_test()
test_labels = prep.data_store.get_y_test(True)
if num_labels == 1:
train_labels = train_labels[:, None]
valid_labels = valid_labels[:, None]
test_labels = test_labels[:, None]
log.debug('shape of train_labels: %s' % (train_labels.shape, ))
graph = tf.Graph()
with graph.as_default():
tf_train_dataset = tf.placeholder(tf.float32,
shape=(batch_size, num_features))
tf_train_labels = tf.placeholder(tf.float32,
shape=(batch_size, num_labels))
tf_valid_dataset = tf.constant(valid_dataset)
tf_test_dataset = tf.constant(test_dataset)
weights_l1 = tf.Variable(
tf.truncated_normal([num_features, hidden_size]))
biases_l1 = tf.Variable(tf.zeros([hidden_size]))
logits_l1 = tf.matmul(tf_train_dataset, weights_l1) + biases_l1
activation_l1 = tf.nn.relu(logits_l1)
weights_l2 = tf.Variable(
tf.truncated_normal([hidden_size, num_labels]))
biases_l2 = tf.Variable(tf.zeros([num_labels]))
logits_l2 = tf.matmul(activation_l1, weights_l2) + biases_l2
if num_labels > 1:
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(
logits_l2, tf_train_labels))
optimizer = tf.train.GradientDescentOptimizer(0.2).minimize(
loss)
# Predictions for the training, validation, and test data
train_prediction = tf.nn.softmax(logits_l2)
valid_layer1 = tf.matmul(tf_valid_dataset,
weights_l1) + biases_l1
valid_prediction = tf.nn.softmax(
tf.matmul(tf.nn.relu(valid_layer1), weights_l2) +
biases_l2)
test_layer1 = tf.matmul(tf_test_dataset,
weights_l1) + biases_l1
test_prediction = tf.nn.softmax(
tf.matmul(tf.nn.relu(test_layer1), weights_l2) + biases_l2)
else:
y_conv = tf.nn.sigmoid(logits_l2)
loss = -(tf_train_labels * tf.log(y_conv + 1e-12) +
(1 - tf_train_labels) * tf.log(1 - y_conv + 1e-12))
loss = tf.reduce_mean(loss)
optimizer = tf.train.GradientDescentOptimizer(0.2).minimize(
loss)
# Predictions for the training data
train_prediction = tf.greater(y_conv, 0.5)
# Predictions for the validation data
valid_layer1 = tf.matmul(tf_valid_dataset,
weights_l1) + biases_l1
valid_layer2 = tf.matmul(tf.nn.relu(valid_layer1),
weights_l2) + biases_l2
y_conv = tf.nn.sigmoid(valid_layer2)
valid_prediction = tf.greater(y_conv, 0.5)
# Predictions for the test data
test_layer1 = tf.matmul(tf_test_dataset,
weights_l1) + biases_l1
test_layer2 = tf.matmul(tf.nn.relu(test_layer1),
weights_l2) + biases_l2
y_conv = tf.nn.sigmoid(test_layer2)
test_prediction = tf.greater(y_conv, 0.5)
num_steps = prep.data_store.get_step_size()
with tf.Session(graph=graph) as session:
tf.initialize_all_variables().run()
print('Initialized')
for step in range(num_steps):
offset = (step * batch_size) % (train_labels.shape[0] -
batch_size)
batch_data = train_dataset[offset:(offset + batch_size), :]
# log.debug('offset: %d, step: %d, shape: %s' % (offset, step, batch_data.shape))
batch_labels = train_labels[offset:(offset + batch_size), :]
feed_dict = {
tf_train_dataset: batch_data,
tf_train_labels: batch_labels
}
_, l, predictions = session.run(
[optimizer, loss, train_prediction], feed_dict=feed_dict)
accuracy = self.get_accuracy_func(num_labels)
if (step % 200 == 0):
print('Minibatch loss at step: %d: %f' % (step, l))
print('Minibatch accuracy: %.1f%%' %
accuracy(predictions, batch_labels))
print('Validation accuracy: %.1f%%' %
accuracy(valid_prediction.eval(), valid_labels))
print('Test accuracy: %.1f%%' %
accuracy(test_prediction.eval(), test_labels))
def get_titanic_data():
global dataset, labels
prep.preprocess_data(force_process=True)
dataset, labels = prep.data_store.get_original_data()
return dataset, labels
def get_dataset(self):
pk.preprocess_data()
self.X_train = pk.data_store.get_X_train(include_valid=True)
self.y_train = pk.data_store.get_y_train(include_valid=True)
self.X_test = pk.data_store.get_X_test()
self.y_test = pk.data_store.get_y_test()