Bdtjtk

I am trying to train my model for Instrument Detection. The output is displaying as loss: nan from the first epoch. I tried to change the loss function, activation function, and add some regularisation like Dropout, but it didn't affect the result.

Here is the code:

from keras.utils import to_categorical

from keras.models import Sequential

from keras.layers import Dense, Conv2D, Flatten, Dropout

from keras.optimizers import Adam



import pickle

import os

import numpy as np

from sklearn.model_selection import train_test_split



def one_hot_encoding(target):

    Instruments = ['vio', 'pia', 'tru', 'flu']

    enc_tar = np.zeros([len(target), 4])



    for i in range(len(target)):

        enc_tar[i][Instruments.index(target[i])] = 1



    return enc_tar



def create_model_cnn(inp_shape):

    classifier = Sequential()



    classifier.add(Conv2D(25, kernel_size = 3, activation = 'relu', input_shape = inp_shape))

    classifier.add(Conv2D(10, kernel_size = 3, activation = 'relu'))

    classifier.add(Flatten())

    classifier.add(Dense(4, activation = 'softmax'))



    adam = Adam(0.001)



    classifier.compile(optimizer = adam, loss = 'categorical_crossentropy', metrics = ['accuracy'])



    return classifier



def create_model_mlp(inp_shape):

    classifier = Sequential()



    classifier.add(Dense(22, activation = 'softsign', input_shape = (42,)))

    classifier.add(Dropout(0.25))

    classifier.add(Dense(10, activation = 'softsign'))

    classifier.add(Dropout(0.25))

    classifier.add(Dense(4, activation = 'softmax'))



    adam = Adam(0.0001)



    classifier.compile(optimizer = adam, loss = 'categorical_crossentropy', metrics = ['accuracy'])



    return classifier



def get_weights(classifier):

    return classifier.get_weights()



def set_weights(classifier, weights):

    classifier.set_weights(weights)

    return classifier



def train_model(classifier, data, target, epoch = 40):

    classifier.fit(data, target, epochs = epoch, validation_split=0.4, batch_size = 32, verbose = 1)

    return classifier



def predict(classifier, data):

    return classifier.predict(data)



if __name__ == '__main__':



    #Get the data and the target

    [data, target] = pickle.load(open('../input/music-features/feat_targ.pickle', 'rb'))



    #if 'model.pickle' not in os.listdir():

        #Generate the classifiers

    cnn_classifier = create_model_cnn((6, 7, 1))

    mlp_classifier = create_model_mlp((42))

    #     else:

    #         #Load the existing model (from a pickle dump)

    #         classifier = pickle.load(open('model.pickle', 'rb'))



    tr_data, tst_data, tr_target, tst_target = train_test_split(data, target)



    tr_data_lin = np.array(tr_data)

    tr_data = tr_data_lin.reshape((tr_data_lin.shape[0], 6, 7, 1))



    tst_data_lin = np.array(tst_data)

    tst_data = tst_data_lin.reshape((tst_data_lin.shape[0], 6, 7, 1))



    enc_target = one_hot_encoding(tr_target)



    #print(tr_data, enc_target)

#     train_model(cnn_classifier, tr_data, enc_target)

    train_model(mlp_classifier, tr_data_lin, enc_target)



#    pickle.dump([cnn_classifier, mlp_classifier], open('model.pickle', 'wb'))

The training data and the test data are from the pickle file where the shape is (15000, 42).