Python keras - getting error with model layers

Question

I am new on classification problems using artificial neural networks

I have a classification problem where the input data are 8 columns with decimal values Which are measures and the output data are 8 columns with integer values which are objects

INPUTS

785.39 6.30 782.75 771.82 7.53 -94.86 378.66 771.82
.
.
.

OUTPUTS

  8     9     5      7     3     1      6      2
  .
  .
  .

The records for the training data are 800 and for the test data are 200

This is the code

import numpy
from keras.models import Sequential
from keras.layers import Dense
from keras.utils import np_utils

seed = 7
numpy.random.seed(seed)
datasetTrain = numpy.loadtxt("train.csv", delimiter=",")
datasetTest = numpy.loadtxt("test.csv", delimiter=",")

X_train = datasetTrain[:,0:7]
y_train = datasetTrain[:,8:15]

X_test = datasetTest[:,0:7]
y_test = datasetTest[:,8:15]

y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)

def baseline_model():
    # create model
    model = Sequential()
    model.add(Dense(7, input_dim=7, kernel_initializer='normal', activation='relu'))
    model.add(Dense((5593, 785), kernel_initializer='normal', activation='softmax'))
    # Compile model
    model.compile(loss='categorical_crossentropy', optimizer='adam')
    return model

model = baseline_model()
model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=400, 
batch_size=200, verbose=25)
scores = model.evaluate(X_test, y_test, verbose=0)
print("Baseline Error: %.2f%%" % (100-scores[1]*100))

I get this error

Traceback (most recent call last):
  File "proyecto.py", line 29, in <module>
    model = baseline_model()
  File "proyecto.py", line 24, in baseline_model
    model.add(Dense((5593, 785), kernel_initializer='normal', activation='softmax'))
ValueError: setting an array element with a sequence.

Which is the best model for this data?

if one of these answers solved your problem please accept it by clicking the check mark next to it. — Max Power
– Max Power, Commented May 15, 2017 at 13:55

Dr. Snoopy · Accepted Answer · 2017-05-12 21:22:09Z

1

This part:

model.add(Dense((5593, 785), kernel_initializer='normal', activation='softmax'))

Is wrong, the first parameter to Dense is the number of output neurons, which should be a scalar, not a tuple or vector. If you want a 2D-shaped output, then you can use a Reshape layer to reshape the output and do the following:

model.add(Dense(5593 * 785, kernel_initializer='normal', activation='softmax'))
model.add(Reshape((5593, 785)))

answered May 12, 2017 at 21:22

Dr. Snoopy

56.5k7 gold badges148 silver badges161 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

Max Power · Accepted Answer · 2017-05-12 18:39:02Z

0

After your first layer model.add(Dense(7..., the output of that layer has dimension 7 (7 output neurons). The next layer can handle that 7-neuron layer as input automatically. But you are then telling Keras that the next layer should have output neurons of (5593, 785) when it is looking for another single number.

Do you get what you want by changing model.add(Dense((5593, 785)...

to either

model.add(Dense(1...

or

model.add(Dense(n...

(where n is the number of possible levels of the categorical variable you're trying to classify?

answered May 12, 2017 at 18:39

Max Power

9,14616 gold badges63 silver badges109 bronze badges

Comments

thefifthjack005 · Accepted Answer · 2017-05-14 11:48:20Z

0

model = Sequential()

num_pixels=img_height*img_width

model.add(Dense(num_pixels, input_dim=num_pixels, kernel_initializer='normal', 
activation='relu'))
model.add(Dense(num_classes, kernel_initializer='normal', 
activation='softmax'))

answered May 14, 2017 at 11:48

thefifthjack005

6384 silver badges9 bronze badges

Collectives™ on Stack Overflow

Python keras - getting error with model layers

3 Answers 3

Comments

Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related