Data Visulization

Data visualization is one of the important processes for data analysis. Because sometimes you may need to see whether your tuning goes to the right direction rather than wrong directions which the result looks more blur and abstract. Also, Visulization can help you demostrate your work to client and show your effort to your boss, as we human are visual animals.

Personally , I did seldom visulizations as i think they are too exaggerating and abstract which give few help to my work. But I will still show something in this article .

I will take the example from last article "cnn to train digit recognition" to visulise the convolution kernel and some output there. so the codes below can be appended after the last block following that article.

1. Kernel shape

# visualize weights
import numpy as np
W = model.layers[0].kernel.get_value(borrow=True)
W = np.squeeze(W)
print ("W shape : " , W.shape)

W shape :  (3, 3, 32)

W[:,:,0] # print the content of 1st convolution kernel

array([[ 0.04594247,  0.03799061, -0.1205595 ],
       [ 0.11602765, -0.08880974,  0.04387789],
       [-0.06434739,  0.0857822 ,  0.09397142]], dtype=float32)

2. Visualize the convolution kernel

# draw out the convolution kernel
%matplotlib inline
import matplotlib.pyplot as plt

for i in range(32):
    sub = plt.subplot(4,8,i+1) # row 4, column 8, ith image 
    plt.axis('off')
    sub.imshow(W[:,:,i], cmap=plt.cm.gray)

3. Print the shape after image is convolved by kernel

# output after convolution
import theano

convout1_f = theano.function(model.inputs,[convout1.output])
C1 = convout1_f([x_train[0]])
C1 = np.squeeze(C1)
print("C1 Shape : ", C1.shape)

C1 Shape :  (6, 6, 32)

Here you may need to install the theano by the way.

4. Visualize image after convolved by the kernel

%matplotlib inline
import matplotlib.pyplot as plt

for i in range(32):
    sub = plt.subplot(4,8,i+1)
    plt.axis('off')
    sub.imshow(C1[:,:,i], cmap=plt.cm.gray)