def predict(test_data, test_labels):
predict_input_fn = tf.estimator.inputs.numpy_input_fn(x={"x": test_data},
num_epochs=1,
shuffle=False)
result = cnn_classifier.predict(predict_input_fn)
#print("result:", list(result))
predict_labels = []
for l in list(result):
#print(l['classes'])
predict_labels.append(l['classes'])
helper.evaluation(predict_labels, test_labels)
def decisionTree():
#x, y = help.generateData()
#train_x, test_x, train_y, test_y = train_test_split(x, y, test_size = 0.33)
train_x, test_x, train_y, test_y = help.generateTrainTest(preload=False,
win_size=8)
## use decisionTree to classify dataset
classifier = tree.DecisionTreeClassifier(criterion='gini')
classifier.fit(train_x, train_y)
joblib.dump(classifier, 'decisionTree.pkl')
classifier = joblib.load('decisionTree.pkl')
predictions = classifier.predict(test_x)
help.evaluation(predictions, test_y)
def DNN():
# get data
# x, y = help.generateData()
# train_x, test_x, train_y, test_y = train_test_split(x, y, test_size = 0.33, random_state = 42)
train_x, test_x, train_y, test_y = help.generateTrainTest(preload=False, win_size=8)
train_y = train_y.astype(int)
test_y = test_y.astype(int)
feature_columns = [tf.contrib.layers.real_valued_column("", dimension = 1440)]
# three 1000-units hidden layers
classifier = tf.contrib.learn.DNNClassifier(feature_columns=feature_columns, hidden_units=[100, 100], n_classes=4)
classifier.fit(x=train_x, y=train_y, steps=2000, batch_size=100)
predictions = list(classifier.predict(test_x, as_iterable=True))
help.evaluation(predictions, test_y)
def LogisticClassifier():
#x, y = help.generateData()
#train_x, test_x, train_y, test_y = train_test_split(x, y, test_size = 0.33, random_state = 42)
train_x, test_x, train_y, test_y = help.generateTrainTest(preload=False,
win_size=0.6)
## use logistic regression to classify dataset
classifier = LogisticRegression(class_weight='balanced',
multi_class='multinomial',
solver='lbfgs')
classifier.fit(train_x, train_y)
joblib.dump(classifier, 'logistic.pkl')
classifier = joblib.load('logistic.pkl')
predictions = classifier.predict(test_x)
help.evaluation(predictions, test_y)
def SVM():
#x, y = help.generateData()
#train_x, test_x, train_y, test_y = train_test_split(x, y, test_size = 0.33, random_state = 42)
train_x, test_x, train_y, test_y = help.generateTrainTest(preload=False,
win_size=8)
## use SVM to classify dataset
# poly is the best kernal for this case, ovr means one vs rest
classifier = svm.SVC(C=0.8,
kernel='poly',
gamma=20,
decision_function_shape='ovr')
classifier.fit(train_x, train_y)
joblib.dump(classifier, 'svm.pkl')
classifier = joblib.load('svm.pkl')
predictions = classifier.predict(test_x)
help.evaluation(predictions, test_y)
step = 0
if sys.argv[1] == "-train":
while step < max_steps:
sess.run(
[train_op],
feed_dict={
x: train_x[step % train_x.shape[0]],
y: train_y[step % train_y.shape[0]],
})
# every 20 steps make a test and save the current model
if step % 20 == 0:
print(sess.run(accuracy, feed_dict={
x: test_x,
y: test_y,
}))
saver.save(sess,
'./model/rnn/my.ckpt',
global_step=tf.train.get_global_step())
step += 1
elif sys.argv[1] == "-test":
help.evaluation(
sess.run(outputlabel, feed_dict={
x: test_x,
y: test_y,
}), test_y.reshape(-1))