How to fix 'Expected to see 2 array(s), but instead got the following list of 1 arrays'
Question:
I’m training a multi class cnn model. model.fit method works fine but when I use fit_generator method the error in the title occurs.
y_train_age = utils.to_categorical(y_train_age, 117)
y_test_age = utils.to_categorical(y_test_age, 117)
y_train_gender = utils.to_categorical(y_train_gender, 2)
y_test_gender = utils.to_categorical(y_test_gender, 2)
y_train = np.concatenate((y_train_age, y_train_gender), axis=1)
y_test = np.concatenate((y_test_age, y_test_gender), axis=1)
print(x_train.shape)
print(x_test.shape)
print(y_train.shape)
print(y_test.shape)
(15000, 100, 100, 3),(8708, 100, 100, 3),(15000, 119),(8708, 119)
Model:
from keras import layers
from keras.models import Model
from keras.layers import Input, Dense, Activation
from keras.layers import AveragePooling2D, MaxPooling2D, Flatten, Conv2D, ZeroPadding2D
x_input = Input((100,100,3))
x = Conv2D(64, (3,3))(x_input)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Conv2D(64, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Conv2D(128, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Conv2D(256, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Flatten()(x)
x = Dense(64, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(128, activation='relu')(x)
y1 = Dense(117, activation='softmax', name="Age")(x)
y2 = Dense(2, activation='softmax', name="Gender")(x)
model = Model(inputs=x_input, outputs=[y1, y2])
model.compile(loss=['categorical_crossentropy', 'categorical_crossentropy'], optimizer='adam', metrics=['accuracy'])
model.summary()
And the problem:
from keras.preprocessing.image import ImageDataGenerator
model.fit_generator(ImageDataGenerator(shear_range=0.3, zoom_range=0.1,
horizontal_flip=True).flow(x_train, y_train, 32),
steps_per_epoch=len(x_train) / 32,
epochs=5, verbose=1,
validation_data=(x_test, y_test))
The error:
ValueError: Error when checking model target: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 2 array(s), but instead got the following list of 1 arrays: [array([[0., 0., 0., ..., 0., 1., 0.],
[0., 0., 0., ..., 0., 0., 1.],
[0., 0., 0., ..., 0., 0., 1.],
...,
[0., 0., 0., ..., 0., 0., 1.],
[0., 0., 0., ..., 0., 0., 1....
Please help me, thanks.
THE ANSWER
generator = ImageDataGenerator(...)
def generate_data_generator(generator, X, Y1, Y2):
genX1 = generator.flow(X, Y1, seed=7)
genX2 = generator.flow(X, Y2, seed=7)
while True:
X1i = genX1.next()
X2i = genX2.next()
yield X1i[0], [X1i[1], X2i[1]]
history = model.fit_generator(generate_data_generator(generator, x_train, y_train_age, y_train_gender),
steps_per_epoch=len(x_train) / 32,
epochs=5,
verbose=1,
callbacks = callbacks,
validation_data=(x_test, [y_test_age, y_test_gender]))
Answers:
You defined a model with 2 outputs : [y1, y2]
So it expected to be fitted with two differents array of labels, one of size (, 119) and the other one of size (,2) which correspond to your 2 Dense outputs layers.
With the fit function it would look like this :
model.fit( x = X_train,
y = [y_train, y_train_gender],
validation_data=(X_test, [y_test, y_test_gender]),
batch_size = batch_size,
epochs = num_epochs,
verbose = 1)
I’am not very used to ImageDataGenerator, but try something like this :
from keras.preprocessing.image import ImageDataGenerator
model.fit_generator(ImageDataGenerator(shear_range=0.3, zoom_range=0.1,
horizontal_flip=True).flow(x_train, [y_train, y_train_gender], 32),
steps_per_epoch=len(x_train) / 32,
epochs=5, verbose=1,
validation_data=(x_test, [y_test, y_test_gender]))
EDIT
Try this little adaptation of this post :
Keras: How to use fit_generator with multiple outputs of different type
generator = ImageDataGenerator(shear_range=0.3,
zoom_range=0.1,
horizontal_flip=True)
def generate_data_generator(generator, X, Y1, Y2):
genX1 = generator.flow(X, Y1, seed=7)
genX2 = generator.flow(X, Y2, seed=7)
while True:
X1i = genX1.next()
X2i = genX2 .next()
yield X1i[0], [X1i[1], X2i[1]]
model.fit_generator(generate_data_generator(generator, x_train, y_train, y_train_gender),
steps_per_epoch=len(x_train) / 32,
epochs=5,
verbose=1)
Just a slight modification to the existing answer, you are using y_train which is a concatenated vector of both Age and Gender but it should just contain the Age since you already have y_train_gender, which encodes the Gender, I have made changes to few segments of the code to accommodate this
y1 = Dense(117, activation='softmax', name="Age")(x) # and not 119
y2 = Dense(2, activation='softmax', name="Gender")(x)
And just replace y_train with y_train_age in your .fit() and .fit_generator() methods. I that way we you use y1 as the output for Age and y2 as the output for Gender.
I’m training a multi class cnn model. model.fit method works fine but when I use fit_generator method the error in the title occurs.
y_train_age = utils.to_categorical(y_train_age, 117)
y_test_age = utils.to_categorical(y_test_age, 117)
y_train_gender = utils.to_categorical(y_train_gender, 2)
y_test_gender = utils.to_categorical(y_test_gender, 2)
y_train = np.concatenate((y_train_age, y_train_gender), axis=1)
y_test = np.concatenate((y_test_age, y_test_gender), axis=1)
print(x_train.shape)
print(x_test.shape)
print(y_train.shape)
print(y_test.shape)
(15000, 100, 100, 3),(8708, 100, 100, 3),(15000, 119),(8708, 119)
Model:
from keras import layers
from keras.models import Model
from keras.layers import Input, Dense, Activation
from keras.layers import AveragePooling2D, MaxPooling2D, Flatten, Conv2D, ZeroPadding2D
x_input = Input((100,100,3))
x = Conv2D(64, (3,3))(x_input)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Conv2D(64, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Conv2D(128, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Conv2D(256, (3,3))(x)
x = Activation('relu')(x)
x = MaxPooling2D((3, 3), strides=(2, 2))(x)
x = Flatten()(x)
x = Dense(64, activation='relu')(x)
x = Dense(128, activation='relu')(x)
x = Dense(128, activation='relu')(x)
y1 = Dense(117, activation='softmax', name="Age")(x)
y2 = Dense(2, activation='softmax', name="Gender")(x)
model = Model(inputs=x_input, outputs=[y1, y2])
model.compile(loss=['categorical_crossentropy', 'categorical_crossentropy'], optimizer='adam', metrics=['accuracy'])
model.summary()
And the problem:
from keras.preprocessing.image import ImageDataGenerator
model.fit_generator(ImageDataGenerator(shear_range=0.3, zoom_range=0.1,
horizontal_flip=True).flow(x_train, y_train, 32),
steps_per_epoch=len(x_train) / 32,
epochs=5, verbose=1,
validation_data=(x_test, y_test))
The error:
ValueError: Error when checking model target: the list of Numpy arrays that you are passing to your model is not the size the model expected. Expected to see 2 array(s), but instead got the following list of 1 arrays: [array([[0., 0., 0., ..., 0., 1., 0.],
[0., 0., 0., ..., 0., 0., 1.],
[0., 0., 0., ..., 0., 0., 1.],
...,
[0., 0., 0., ..., 0., 0., 1.],
[0., 0., 0., ..., 0., 0., 1....
Please help me, thanks.
THE ANSWER
generator = ImageDataGenerator(...)
def generate_data_generator(generator, X, Y1, Y2):
genX1 = generator.flow(X, Y1, seed=7)
genX2 = generator.flow(X, Y2, seed=7)
while True:
X1i = genX1.next()
X2i = genX2.next()
yield X1i[0], [X1i[1], X2i[1]]
history = model.fit_generator(generate_data_generator(generator, x_train, y_train_age, y_train_gender),
steps_per_epoch=len(x_train) / 32,
epochs=5,
verbose=1,
callbacks = callbacks,
validation_data=(x_test, [y_test_age, y_test_gender]))
You defined a model with 2 outputs : [y1, y2]
So it expected to be fitted with two differents array of labels, one of size (, 119) and the other one of size (,2) which correspond to your 2 Dense outputs layers.
With the fit function it would look like this :
model.fit( x = X_train,
y = [y_train, y_train_gender],
validation_data=(X_test, [y_test, y_test_gender]),
batch_size = batch_size,
epochs = num_epochs,
verbose = 1)
I’am not very used to ImageDataGenerator, but try something like this :
from keras.preprocessing.image import ImageDataGenerator
model.fit_generator(ImageDataGenerator(shear_range=0.3, zoom_range=0.1,
horizontal_flip=True).flow(x_train, [y_train, y_train_gender], 32),
steps_per_epoch=len(x_train) / 32,
epochs=5, verbose=1,
validation_data=(x_test, [y_test, y_test_gender]))
EDIT
Try this little adaptation of this post :
Keras: How to use fit_generator with multiple outputs of different type
generator = ImageDataGenerator(shear_range=0.3,
zoom_range=0.1,
horizontal_flip=True)
def generate_data_generator(generator, X, Y1, Y2):
genX1 = generator.flow(X, Y1, seed=7)
genX2 = generator.flow(X, Y2, seed=7)
while True:
X1i = genX1.next()
X2i = genX2 .next()
yield X1i[0], [X1i[1], X2i[1]]
model.fit_generator(generate_data_generator(generator, x_train, y_train, y_train_gender),
steps_per_epoch=len(x_train) / 32,
epochs=5,
verbose=1)
Just a slight modification to the existing answer, you are using y_train which is a concatenated vector of both Age and Gender but it should just contain the Age since you already have y_train_gender, which encodes the Gender, I have made changes to few segments of the code to accommodate this
y1 = Dense(117, activation='softmax', name="Age")(x) # and not 119
y2 = Dense(2, activation='softmax', name="Gender")(x)
And just replace y_train with y_train_age in your .fit() and .fit_generator() methods. I that way we you use y1 as the output for Age and y2 as the output for Gender.