Aggregating output classes into one category in Theano/Keras - deep-learning

Question

I'm building a ConvNet for handwritten digits recognition using the MNIST dataset. My code is written in Keras using the Theano backend.
I want to train my ConvNet so it can identify a subset of classes (for example, digits '1' and '2' only) and output any other as a generic 'unknown' class. I know this can be done on Theano since it's described on "Distributed Neural Networks for Internet of Things: The Big-Little Approach", but I can't find any documentation or examples on this subject.

Answer 1

Modify your objective function outputs (your Y_train) to convert 0, 1, 2, 3, 4, 5, etc., to 0, 1, 2 (where 2 represents "other"). Note that if you actually want to predict, say, digits 5 and 6, and everything else as other, you'll need to re-index your classes to start at 0, so that 0 becomes digit 5, 1 becomes digit 6, and 2 becomes "other".
Here's a modified version of the Keras MNIST example, basically straight from the repo with just a few extra lines:
'''Trains a simple convnet on the MNIST dataset.
Gets to 99.25% test accuracy after 12 epochs
(there is still a lot of margin for parameter tuning).
16 seconds per epoch on a GRID K520 GPU.
'''
from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
import numpy as np
batch_size = 128
epochs = 1
# input image dimensions
img_rows, img_cols = 28, 28
# the data, shuffled and split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()
###############################
# This is the key... order is important!
y_train[y_train<=4]=2
y_train[y_train==5]=0
y_train[y_train==6]=1
y_train[y_train>=7]=2
y_test[y_test<=4]=2
y_test[y_test==5]=0
y_test[y_test==6]=1
y_test[y_test>=7]=2
num_classes=3
print(np.unique(y_train))
# [0 1 2]
###############################
if K.image_data_format() == 'channels_first':
x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
input_shape = (1, img_rows, img_cols)
else:
x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
activation='relu',
input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adadelta(),
metrics=['accuracy'])
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])