classification mnist binary class

Question

I want to implement a Neural Network for MNIST dataset, and I find a lot of example on the net. But I want study this problem using a different approach: I want create 10 NNs (as the number of classes) in which I classify only a class with the rest of the others (example: first NN analyze only the "1" class vs the others). It's just an exercise (I'm python newbie and I want to learn).

This is my actually python code, do you have any suggests to modify mi code in order to obtain the classes separation that I explained above?

from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense
import numpy
from keras.utils import np_utils
numpy.random.seed()


(X_train, y_train), (X_test, y_test) = mnist.load_data()

num_pixels = X_train.shape[1] * X_train.shape[2]
X_train.shape[1],X_train.shape[2],X_train.shape[3])
X_train = X_train.reshape(X_train.shape[0], num_pixels).astype('float32')
X_test = X_test.reshape(X_test.shape[0], num_pixels).astype('float32')

X_train = X_train / 255
X_test = X_test / 255
y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)
print(y_test)
num_classes = y_test.shape[1]
def baseline_model():
model = Sequential()
model.add(Dense(num_pixels, input_dim=num_pixels,           kernel_initializer='normal', activation='relu'))
model.add(Dense(num_classes, kernel_initializer='normal', activation='softmax'))        

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


 model = baseline_model()

 model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=3,     batch_size=200, verbose=2)

 model.summary()
 scores = model.evaluate(X_test, y_test, verbose=1)
 print("Baseline Error: %.2f%%" % (100-scores[1]*100))

Imran · Accepted Answer · 2018-02-18 21:42:38Z

1

First you should transform y_train and y_test into binary outputs. You will want to do this in a loop for each of the ten classes, but here is how you would do it for one of those, say 5:

y_train_binary = (y_train == 5).astype(np.int)
y_test_binary = (y_test == 5).astype(np.int)

Right now each y_train example is a number from 1 to 10, so here we convert it to 0 or 1 corresponding to whether the positive class is correct.

Then you want to adjust the output of your model to support binary classification. This can be done by using a single dense output neuron with sigmoid activation and binary crossentropy loss:

model.add(Dense(1, kernel_initializer='normal', activation='sigmoid'))   
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])

edited Feb 18, 2018 at 21:42

answered Feb 18, 2018 at 9:32

Imran

13.6k8 gold badges69 silver badges82 bronze badges

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

12 Comments

7speed Over a year ago

Thank you very much Imran! i dont't understand why I use 5...can you explain me? Furthermore, np.int was marked as "undefined", I must import any other library? Sorry for my bad question, I'm a newbie for Python

Imran Over a year ago

That was just an example - each entry in y_train is a number between 1 and 10. In my example we convert 5 to array([0, 0, 0, 0, 1, 0, 0, 0, 0, 0]). You should wrap everything below load_data() in a loop, outputting a different model for each class.

7speed Over a year ago

Thanks, and for the np.int?

Imran Over a year ago

(y_train == 5) gives you an array of booleans, and astype(np.int) converts it to 1s and 0s.

7speed Over a year ago

Perfect I understand, but my editor mark np as undefined. Why? It's necessary import a library?

|

Collectives™ on Stack Overflow

classification mnist binary class

1 Answer 1

12 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

12 Comments

Your Answer

Sign up or log in

Post as a guest

Related