Tensorflow Convolutional Neural Network

A basic implementation of an artificial neural network on python using numpy, matplotlib and tensorflow.

The dataset

The dataset used for this project is the mnist dataset. It consists of 60000, 28 x 28 samples of handwritten digits which are used to train the network. A sample from the dataset can be visualized as follows

from tensorflow.examples.tutorials.mnist import input_data
data = input_data.read_data_sets('/tmp/data', one_hot = True)

Xdata = data.test.images
sample = Xdata[741].reshape(28,28)
plt.imshow(sample, cmap='Greys')

Further information about this dataset is available on Kaggle.

The problem

The main goal of this neural network is to predict handwitten digits with the highest possible accuracy, for this we will use a convolutional neural network consisting of 2 convolutional layers (with reLU as activation function), and one hidden layer for the fully conected layer and a 0.1 dropout rate for that layer.

Results

After training for 30 epochs we observe that the training and validation costs decrease as expected obtain an accuracy of aproximately 98.5 which does not increase majorly if we increase the epoch number.

We can check that it also predicts digits that are not contained on the dataset itself, for this, a sample folder is included that contains digits made on MS Paint.

from skimage import color
from skimage import io
custom_photo = color.rgb2gray(io.imread('samples/three.jpg'))
custom_photo = np.array(custom_photo)
custom_photo = np.float32(custom_photo)
custom_photo = 1 - custom_photo
custom_photo[custom_photo < 0.05] = 0 

with tf.Session() as sess:
        sess.run(init_op)
        saver.restore(sess, "model.ckpt")
        prediction=sess.run(conv_neural_network(custom_photo, mode='p'))
print('The number is:', np.argmax(prediction))

And it outputs

The number is: 3