unsupported operand type error when using Keras interface on tensorflow code for CNN's

[shksa]

``

In[1]:

from six.moves import cPickle as pickle
from six.moves import range
import numpy as np
from PIL import Image
import os
import sys
import matplotlib.pyplot as plt

In[ ]:

pickle_file = 'SVHN_multi_normal.pickle'
with open(pickle_file, 'rb') as f:
save = pickle.load(f)
train_dataset = save['train_dataset_n']
train_labels = save['train_labels_n']
test_dataset = save['test_dataset_n']
test_labels = save['test_labels_n']
del save # hint to help gc free up memory
print('Training set', train_dataset.shape, train_labels.shape)
print('Test set', test_dataset.shape, test_labels.shape)

train_dataset = train_dataset.astype('float32')
test_dataset = test_dataset.astype('float32')
train_dataset = train_dataset / 255.0
test_dataset = test_dataset / 255.0

In[ ]:

import tensorflow as tf
from keras.models import Sequential
from keras.layers import Convolution2D, MaxPooling2D, Flatten, Dense, Activation, Dropout, BatchNormalization
from keras.optimizers import Adam

In[ ]:

def accuracy(predictions, labels):
return (100.0 * np.sum(np.argmax(predictions, 2).T == labels) / predictions.shape[1] / predictions.shape[0])

batch_size = 64

In[ ]:

my_f_session = tf.Session()
from keras import backend as K
K.set_session(my_f_session)

tf_train_data = tf.placeholder(tf.float32, shape=(None, 54, 54, 3))
tf_train_labels = tf.placeholder(tf.int32, shape=(None, 6))
tf_test_dataset = tf.constant(test_dataset)

def my_model(input_batch):

# Block 1
x = Convolution2D(32, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block1_conv1')(input_batch)
x = BatchNormalization(name="block1_bn1")(x)
x = Convolution2D(32, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block1_conv2')(x)
x = BatchNormalization(name="block1_bn2")(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)
x = Dropout(0.3)(x)

# Block 2
x = Convolution2D(64, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block2_conv1')(x)
x = BatchNormalization(name="block2_bn1")(x)
x = Convolution2D(64, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block2_conv2')(x)
x = BatchNormalization(name="block2_bn2")(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)
x = Dropout(0.3)(x)

# Block 3
x = Convolution2D(128, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block3_conv1')(x)
x = BatchNormalization(name="block3_bn1")(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block3_pool')(x)
x = Dropout(0.3)(x)

# Block 4
x = Convolution2D(256, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block4_conv1')(x)
x = BatchNormalization(name="block4_bn1")(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block4_pool')(x)
x = Dropout(0.3)(x)

# Block 5
x = Convolution2D(512, 3, 3, activation='relu', init='he_normal', border_mode='same', name='block5_conv1')(x)
x = BatchNormalization(name="block5_bn1")(x)
x = MaxPooling2D((2, 2), strides=(2, 2), name='block5_pool')(x)

f = Flatten(name='flatten')(x)
h = Dense(4096, activation='relu', init='he_normal')(f)

logits1 = Dense(11, init='he_normal')(h)
logits2 = Dense(11, init='he_normal')(h)
logits3 = Dense(11, init='he_normal')(h)
logits4 = Dense(11, init='he_normal')(h)
logits5 = Dense(11, init='he_normal')(h)

return [logits1, logits2, logits3, logits4, logits5]

[logits1, logits2, logits3, logits4, logits5] = my_model(tf_train_data)

In[ ]:

loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits1, tf_train_labels[:,1])) + tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits2, tf_train_labels[:,2])) + tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits3, tf_train_labels[:,3])) + tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits4, tf_train_labels[:,4])) + tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits5, tf_train_labels[:,5]))

In[ ]:

train_step = tf.train.AdamOptimizer().minimize(loss)

In[ ]:

train_prediction = tf.pack([tf.nn.softmax(my_model(tf_train_data)[0]),
tf.nn.softmax(my_model(tf_train_data)[1]),
tf.nn.softmax(my_model(tf_train_data)[2]),
tf.nn.softmax(my_model(tf_train_data)[3]),
tf.nn.softmax(my_model(tf_train_data)[4])])

test_prediction = tf.pack([tf.nn.softmax(my_model(tf_test_dataset)[0]),
tf.nn.softmax(my_model(tf_test_dataset)[1]),
tf.nn.softmax(my_model(tf_test_dataset)[2]),
tf.nn.softmax(my_model(tf_test_dataset)[3]),
tf.nn.softmax(my_model(tf_test_dataset)[4])])

In[ ]:

num_steps = 100001
with my_f_session.as_default():

K.get_session().run(tf.global_variables_initializer())
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), :, :, :]
    batch_labels = train_labels[offset:(offset + batch_size),:]
    feed_dict = {tf_train_data : batch_data, tf_train_labels : batch_labels, K.learning_phase(): 1, K.learning_phase(): 1}
    _, l = my_f_session.run([train_step, loss], feed_dict=feed_dict)
    if (step % 500 == 0): 
        print('Minibatch loss at step %d: %f' % (step, l))
        feed_dict = {tf_train_data : batch_data, K.learning_phase(): 0, K.learning_phase(): 0}
        predictions = my_f_session.run([train_prediction], feed_dict=feed_dict)
        print('Minibatch accuracy: %.1f%%' % accuracy(predictions, batch_labels[:,1:6]))
feed_dict = {K.learning_phase(): 0, K.learning_phase(): 0}
predictions = my_f_session.run([test_prediction], feed_dict=feed_dict)
print('Test accuracy: %.1f%%' % accuracy(predictions, test_labels[:,1:6]))

File "C:\Users\sreekar\Desktop\UdacityProject\hangyao\5_CNN_multi_my_version.py", line 106, in
[logits1, logits2, logits3, logits4, logits5] = my_model(tf_train_data)
File "C:\Users\sreekar\Desktop\UdacityProject\hangyao\5_CNN_multi_my_version.py", line 96, in my_model
h = Dense(4096, activation='relu', init='he_normal')(f)
File "C:\Program Files\Python35\lib\site-packages\keras\engine\topology.py", line 546, in call
self.build(input_shapes[0])
File "C:\Program Files\Python35\lib\site-packages\keras\layers\core.py", line 798, in build
constraint=self.W_constraint)
File "C:\Program Files\Python35\lib\site-packages\keras\engine\topology.py", line 418, in add_weight
weight = initializer(shape, name=name)
File "C:\Program Files\Python35\lib\site-packages\keras\initializations.py", line 76, in he_normal
s = np.sqrt(2. / fan_in)
TypeError: unsupported operand type(s) for /: 'float' and 'NoneType'

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed after 30 days if no further activity occurs, but feel free to re-open a closed issue if needed.

[aditipanda]

did anyone solve this?