IndexError: list index out of range when doing model.fit on keras

[emnaj...@gmail.com]

Hi everyone,

I have written my own custom Loss function. I wanted to be calculated per batch. 

def calculate_additional_loss(y_true,y_pred):

    #additional loss

    x_decoded_normalized = tf.round(original_dim* y_pred)

    error = tf.constant(0, dtype= tf.float32)

    additional_loss= tf.constant(0, dtype= tf.float32)

    final_loss = 0 * mse(inputs,outputs)

        for k in range(batch_size):

        #add padding

        reshaped_elem_1 = K.reshape(x_decoded_normalized[k], [DIM,DIM])

        

        a = K.reshape(reshaped_elem_1[:,DIM-1], [DIM,1])

        b = K.reshape(reshaped_elem_1[:,1], [DIM,1])

        reshaped_elem_1 = tf.concat ([b,reshaped_elem_1], axis= 1)

        reshaped_elem_1 = tf.concat ([reshaped_elem_1,a], axis= 1)

        #print('here', reshaped_elem_1.shape)

        

        c= K.reshape(reshaped_elem_1[DIM-1,:], [1,DIM+2])

        d= K.reshape(reshaped_elem_1[1,:], [1,DIM+2])

        reshaped_elem_1 = tf.concat ([d,reshaped_elem_1],axis=0)

        reshaped_elem_1 = tf.concat ([reshaped_elem_1,c],axis=0)

        for (i,j) in range(reshaped_elem_1.shape[0],reshaped_elem_1.shape[1]):

            error = tf.add(error, tf.pow((reshaped_elem_1[i,j]-reshaped_elem_1[i,j+1]),-2), tf.pow((reshaped_elem_1[i,j]-reshaped_elem_1[i,j-1]),-2), tf.pow((reshaped_elem_1[i,j]-reshaped_elem_1[i-1,j]),-2), tf.pow((reshaped_elem_1[i,j]-reshaped_elem_1[i+1,j]),-2))

                        #print('error', error)

        #print('additional_loss', K.get_value(additional_loss))

        additional_loss = tf.add(additional_loss, tf.divide(error, original_dim))

      final_loss += tf.divide(additional_loss, batch_size)

    #sess.close()

    return final_loss

For training the model, I am using this code :

sess = tf.Session()

K.set_session(sess)

K.set_learning_phase(0)

models = (encoder, decoder)

vae.compile(loss=[calculate_additional_loss], optimizer='adam', metrics= ['acc'])

vae.summary()

plot_model(vae,to_file='vae_mlp.png',show_shapes=True)

vae.fit(x_train, epochs=epochs, batch_size=batch_size, validation_data=(x_test, None), verbose = 1, callbacks=[CustomMetrics()])

vae.save_weights('vae_mlp_mnist.h5')

My model is variational autoencoder. The encoder architecture is this one :

The decoder architecture is this one:

The error I am getting while running the model is :

Train on 20060 samples, validate on 2000 samples

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-84-9ba5be4c3154> in <module>()
     36 
     37 plot_model(vae,to_file='vae_mlp.png',show_shapes=True)
---> 38 vae.fit(x_train, epochs=epochs, batch_size=batch_size, validation_data=(x_test, None), verbose = 1)
     39 #metrics = vae.evaluate(x=x_train, y=None, batch_size=batch_size, verbose=1)
     40 #print(metrics)

/opt/aiml4it/anaconda/3-5.2.0-generic/lib/python3.6/site-packages/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)
   1040                                         initial_epoch=initial_epoch,
   1041                                         steps_per_epoch=steps_per_epoch,
-> 1042                                         validation_steps=validation_steps)
   1043 
   1044     def evaluate(self, x=None, y=None,

/opt/aiml4it/anaconda/3-5.2.0-generic/lib/python3.6/site-packages/keras/engine/training_arrays.py in fit_loop(model, f, ins, out_labels, batch_size, epochs, verbose, callbacks, val_f, val_ins, shuffle, callback_metrics, initial_epoch, steps_per_epoch, validation_steps)
    135     indices_for_conversion_to_dense = []
    136     for i in range(len(feed)):
--> 137         if issparse(ins[i]) and not K.is_sparse(feed[i]):
    138             indices_for_conversion_to_dense.append(i)
    139 

IndexError: list index out of range

I cannot find an explanation for this error. It doesn't show as well in which line. Does anyone have an idea about it please ?

Thank you in advance.

[jeetu...@gmail.com]

Any progress with this ??. I am also getting error for this same thing

[Ted Yu]

Is it possible to show your encoder / decoder code ?

Thanks

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/37b3ddf4-82f6-42c3-aec4-45a53d647b60%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[emnaj...@gmail.com]

I tried to rewrite the additional loss function and it worked out for me.

This is the new code.

i = tf.Variable(0)

k = tf.Variable(0)

l = tf.Variable(0)

error = tf.Variable(initial_value=0,dtype=tf.float32)

error_1 = tf.Variable(initial_value=0,dtype=tf.float32)

one = tf.constant(1, dtype=tf.float32)

final_loss = tf.Variable(0)

def cond_1(k,l, sol, error):

    return tf.logical_and((tf.less(k, 13)),(tf.less(l, 13)))

def body_1(k,l, sol,error):

    elem_1 = tf.abs(sol[k,l]-sol[k,l+1])

    elem_2 = tf.abs(sol[k,l]-sol[k,l-1])

    sum_1 = tf.add(elem_1, elem_2)

    elem_3 = tf.abs(sol[k,l]-sol[k-1,l])

    elem_4 = tf.abs(sol[k,l]-sol[k+1,l])

    sum_2 = tf.add(sum_1, elem_3)

    sum_3 = tf.add(sum_2,elem_4)

    error = tf.add(error,tf.add(sum_3,one))

    k = tf.add(k, 1)

    l = tf.add(l, 1)

    return [k,l, sol, error] 

def cond(i, sol, error_1):

    return tf.less(i, 10)

def body(i, sol,error_1):

    

    original_reshaped_elem = sol[i]

    original_reshaped_elem = tf.reshape(original_reshaped_elem, [DIM,DIM])

    

    a = tf.reshape(original_reshaped_elem[:,DIM-1], [DIM,1])

    b = tf.reshape(original_reshaped_elem[:,1], [DIM,1])

    original_reshaped_elem = tf.concat ([b,original_reshaped_elem], axis= 1)

    original_reshaped_elem = tf.concat ([original_reshaped_elem,a], axis= 1)

    c= tf.reshape(original_reshaped_elem[DIM-1,:], [1,DIM+2])

    d= tf.reshape(original_reshaped_elem[1,:], [1,DIM+2])

    original_reshaped_elem = tf.concat ([d,original_reshaped_elem],axis=0)

    reshaped_elem_extended = tf.concat ([original_reshaped_elem,c],axis=0)

    result = tf.while_loop(cond_1, body_1, [k,l, reshaped_elem_extended, error])

    error_1 = tf.divide(result[3], 144)

    i = tf.add(i, 1)

    return [i, sol, error_1]

and this is where I am calling it in the keras model

result = tf.while_loop(cond, body, [i, outputs, error_1])

final_loss = tf.divide(result[2], 10)

[emnaj...@gmail.com]

Encoder code :

# build encoder model

inputs = Input(shape=input_shape, name='encoder_input')

x = Dense(intermediate_dim, activation='relu')(inputs)

z_mean = Dense(latent_dim, name='z_mean')(x)

z_log_var = Dense(latent_dim, name='z_log_var')(x)

Decoder part:

latent_inputs = Input(shape=(latent_dim,), name='z_sampling')

x = Dense(intermediate_dim, activation='relu')(latent_inputs)

outputs = Dense(original_dim, activation='relu')(x)