def evaluate(num):
# num 表示要取那个人的数据
# return 第num 个人数据经过测试得到数据。
with tf.name_scope("input"):
input_x = tf.placeholder(tf.float32, [bat[num], 5000, 9, 1],
name='EEG-input') # 数据的输入，第一维表示一个batch中样例的个数
input_y = tf.placeholder(tf.float32, [None, 7],
name='EEG-lable') # 一个batch里的lable
regularlizer = tf.contrib.layers.l2_regularizer(
Regularazition_Rate) #本来测试的时候不用加这个
is_training = tf.cast(False, tf.bool)
out = CNN_LSTM(input_x, is_training, None)
y = out['logist']
with tf.name_scope("test_acc"):
correct_predection = tf.equal(tf.argmax(y, 1), tf.argmax(input_y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_predection, tf.float32))
tf.summary.scalar('test_acc', accuracy)
variable_averages = tf.train.ExponentialMovingAverage(Moving_Average_Decay)
variables_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variables_to_restore)
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(Model_Save_Path)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split('/')[-1].split(
'-')[-1]
x, y = raw_test_batch10(num) #获取第x个人的数据
reshape_xs = np.reshape(x, (-1, 5000, 9, 1))
ys = one_hot(y)
conv1, pool1, conv2, pool2, conv3, pool3, conv4, pool4, lstm, acc_score = sess.run(
[
out['conv1'], out['pool1'], out['conv2'], out['pool2'],
out['conv3'], out['pool3'], out['conv4'], out['pool4'],
out['rnn'], accuracy
],
feed_dict={
input_x: reshape_xs,
input_y: ys
})
pool4 = np.reshape(pool4, (-1, 313, 1, 128))
Lstm = np.reshape(lstm, (-1, 313, 1, 128))
print("Afer %s training step, test accuracy = %g" %
(global_step, acc_score))
else:
print("No checkpoint file found")
return conv1, pool1, conv2, pool2, conv3, pool3, conv4, pool4, Lstm
def evaluate(num):
# num 表示要取那个人的数据
# return 第num 个人数据经过测试得到数据。
with tf.name_scope("input"):
input_x = tf.placeholder(tf.float32, [bat[num], 5000, 9, 1],
name='EEG-input') # 数据的输入，第一维表示一个batch中样例的个数
input_y = tf.placeholder(tf.float32, [None, 7],
name='EEG-lable') # 一个batch里的lable
regularlizer = tf.contrib.layers.l2_regularizer(
Regularazition_Rate) #本来测试的时候不用加这个
is_training = tf.cast(False, tf.bool)
out = BaseCNN(input_x, is_training, regularlizer)
y = out['logit']
with tf.name_scope("test_acc"):
correct_predection = tf.equal(tf.argmax(y, 1), tf.argmax(input_y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_predection, tf.float32))
tf.summary.scalar('test_acc', accuracy)
variable_averages = tf.train.ExponentialMovingAverage(Moving_Average_Decay)
variables_to_restore = variable_averages.variables_to_restore()
with tf.Session() as sess:
tf.initialize_all_variables().run()
x, y = raw_test_batch10(num) #获取第x个人的数据
reshape_xs = np.reshape(x, (-1, 5000, 9, 1))
ys = one_hot(y)
conv1, pool1, conv2, pool2, conv3, pool3, conv4, pool4, acc_score = sess.run(
[
out['conv1'], out['pool1'], out['conv2'], out['pool2'],
out['conv3'], out['pool3'], out['conv4'], out['pool4'],
accuracy
],
feed_dict={
input_x: reshape_xs,
input_y: ys
})
print("Afer training step, test accuracy = %g" % (acc_score))
return conv1, pool1, conv2, pool2, conv3, pool3, conv4, pool4
def elva(data):
x_test, y_test = raw_test_batch10(data)
x_test = sequence.pad_sequences(x_test, maxlen=maxlen, dtype='float32')
model = Sequential()
model.add(
Conv1D(filters, kernel_size, padding='same', strides=1, name='conv1'))
model.add(BatchNormalization())
model.add(Activation('elu'))
model.add(MaxPooling1D(pool_size=pool_size, name='pool1'))
model.add(Conv1D(36, kernel_size, padding='same', strides=1, name='conv2'))
model.add(BatchNormalization())
model.add(Activation('elu'))
model.add(MaxPooling1D(pool_size=pool_size, name='pool2'))
model.add(Conv1D(72, kernel_size, padding='same', strides=1, name='conv3'))
model.add(BatchNormalization())
model.add(Activation('elu'))
model.add(MaxPooling1D(pool_size=pool_size, name='pool3'))
model.add(Conv1D(144, kernel_size, padding='same', strides=1,
name='conv4'))
model.add(BatchNormalization())
model.add(Activation('elu'))
model.add(MaxPooling1D(pool_size=pool_size, name='pool4'))
model.add(LSTM(lstm_output_size, name='lstm'))
model.add(Dense(7, name='dense1'))
model.add(Activation('softmax', name='softmax'))
adam = Adam(0.00325)
model.compile(loss='sparse_categorical_crossentropy',
optimizer=adam,
metrics=['accuracy'])
model = load_model('CNN_LSTM_model_v23.h5')
score, acc = model.evaluate(x_test, y_test)
# model.load_weights("my_model_weights.h5", by_name=True)
# intermediate_layer_model1 = K.function(inputs=[model.layers[0].inputs], outputs=[model.get_layer('conv1').output])
# print(model.inputs)
intermediate_layer_model1 = Model(inputs=model.inputs,
outputs=model.get_layer('conv1').output)
intermediate_layer_model2 = Model(inputs=model.inputs,
outputs=model.get_layer('pool1').output)
intermediate_layer_model3 = Model(inputs=model.inputs,
outputs=model.get_layer('conv2').output)
intermediate_layer_model4 = Model(inputs=model.inputs,
outputs=model.get_layer('pool2').output)
intermediate_layer_model5 = Model(inputs=model.inputs,
outputs=model.get_layer('conv3').output)
intermediate_layer_model6 = Model(inputs=model.inputs,
outputs=model.get_layer('pool3').output)
intermediate_layer_model7 = Model(inputs=model.inputs,
outputs=model.get_layer('conv4').output)
intermediate_layer_model8 = Model(inputs=model.inputs,
outputs=model.get_layer('pool4').output)
intermediate_layer_model9 = Model(inputs=model.inputs,
outputs=model.get_layer('lstm').output)
intermediate_output1 = intermediate_layer_model1.predict(x_test)
print(intermediate_output1.shape)
intermediate_output2 = intermediate_layer_model2.predict(x_test)
print(intermediate_output2.shape)
intermediate_output3 = intermediate_layer_model3.predict(x_test)
print(intermediate_output3.shape)
intermediate_output4 = intermediate_layer_model4.predict(x_test)
print(intermediate_output4.shape)
intermediate_output5 = intermediate_layer_model5.predict(x_test)
print(intermediate_output5.shape)
intermediate_output6 = intermediate_layer_model6.predict(x_test)
print(intermediate_output6.shape)
intermediate_output7 = intermediate_layer_model7.predict(x_test)
print(intermediate_output7.shape)
intermediate_output8 = intermediate_layer_model8.predict(x_test)
print(intermediate_output8.shape)
intermediate_output9 = intermediate_layer_model9.predict(x_test)
print(intermediate_output9.shape)
out = {
'conv1': intermediate_output1,
'pool1': intermediate_output2,
'conv2': intermediate_output3,
'pool2': intermediate_output4,
'conv3': intermediate_output5,
'pool3': intermediate_output6,
'conv4': intermediate_output7,
'pool4': intermediate_output8,
'lstm': intermediate_output9,
}
return out