notebook.community

Edit and run

Autoencoder

Building an autoencoder using Keras.

Update environment





In [1]:



    


!apt-get install graphviz -qq
!pip install graphviz -q
!pip install pydot -q
# restart runtime



    











In [2]:



    


import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

from keras.layers import Input, Dense
from keras.models import Model
from keras.datasets import mnist



    


















    






Using TensorFlow backend.

















In [3]:



    


# copy-paste from ml-stuff/kaggle/utils_keras.py
# delete before saving for brevity
# ...

Creating the model





In [4]:



    


# this is the size of our encoded representations
encoding_dim = 32  # 32 floats -> compression of factor 24.5, assuming the input is 784 floats

# this is our input placeholder
input_img = Input(shape=(784,))

# encoded representation of the input
encoded = Dense(encoding_dim, activation='relu')(input_img)
# lossy reconstruction of the input
decoded = Dense(784, activation='sigmoid')(encoded)

# maps an input to its reconstruction
autoencoder = Model(input_img, decoded)

# maps an input to its encoded representation
encoder = Model(input_img, encoded)


# placeholder for an encoded (32-dimensional) input
encoded_input = Input(shape=(encoding_dim,))

# retrieve the last layer of the autoencoder model
decoder_layer = autoencoder.layers[-1]

# decoder model
decoder = Model(encoded_input, decoder_layer(encoded_input))



    











In [5]:



    


autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')

Training the model





In [6]:



    


(x_train, _), (x_test, _) = mnist.load_data()



    











In [7]:



    


x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.

x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))

print(x_train.shape)
print(x_test.shape)



    


















    






(60000, 784)
(10000, 784)

















In [8]:



    


autoencoder.fit(x_train, x_train,
                epochs=20,
                batch_size=256,
                shuffle=True,
                validation_data=(x_test, x_test))



    


















    






Train on 60000 samples, validate on 10000 samples
Epoch 1/20
60000/60000 [==============================] - 3s 55us/step - loss: 0.3571 - val_loss: 0.2705
Epoch 2/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.2631 - val_loss: 0.2515
Epoch 3/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.2401 - val_loss: 0.2265
Epoch 4/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.2186 - val_loss: 0.2089
Epoch 5/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.2043 - val_loss: 0.1972
Epoch 6/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.1940 - val_loss: 0.1882
Epoch 7/20
14336/60000 [======>.......................] - ETA: 1s - loss: 0.188360000/60000 [==============================] - 2s 35us/step - loss: 0.1858 - val_loss: 0.1807
Epoch 8/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.1789 - val_loss: 0.1745
Epoch 9/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.1731 - val_loss: 0.1691
Epoch 10/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.1679 - val_loss: 0.1643
Epoch 11/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.1632 - val_loss: 0.1596
Epoch 12/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.1590 - val_loss: 0.1555
Epoch 13/20
49664/60000 [=======================>......] - ETA: 0s - loss: 0.155360000/60000 [==============================] - 2s 35us/step - loss: 0.1550 - val_loss: 0.1517
Epoch 14/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.1513 - val_loss: 0.1482
Epoch 15/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.1479 - val_loss: 0.1450
Epoch 16/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.1448 - val_loss: 0.1418
Epoch 17/20
60000/60000 [==============================] - 2s 35us/step - loss: 0.1419 - val_loss: 0.1391
Epoch 18/20
60000/60000 [==============================] - 2s 36us/step - loss: 0.1392 - val_loss: 0.1365
Epoch 19/20
53760/60000 [=========================>....] - ETA: 0s - loss: 0.136860000/60000 [==============================] - 2s 36us/step - loss: 0.1367 - val_loss: 0.1340
Epoch 20/20
60000/60000 [==============================] - 2s 37us/step - loss: 0.1344 - val_loss: 0.1318










    
Out[8]:








<keras.callbacks.History at 0x7f094de029b0>

Testing the model





In [9]:



    


# encode and decode some digits
encoded_imgs = encoder.predict(x_test)
decoded_imgs = decoder.predict(encoded_imgs)



    











In [10]:



    


plt.figure(figsize=(20, 4))

n = 10  # no of digits to display

for i in range(n):
    # display original
    ax = plt.subplot(2, n, i + 1)
    plt.imshow(x_test[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

    # display reconstruction
    ax = plt.subplot(2, n, i + 1 + n)
    plt.imshow(decoded_imgs[i].reshape(28, 28))
    plt.gray()
    ax.get_xaxis().set_visible(False)
    ax.get_yaxis().set_visible(False)

plt.savefig('plot_predict_vs_reconstructed.png')
plt.show()

Save model





In [11]:



    


models = {
	'autoencoder': autoencoder,
	'encoder': encoder,
	'decoder': decoder
}

model_save_mult(models)



    


















    






Saved autoencoder to disk
Saved encoder to disk
Saved decoder to disk

















In [12]:



    


spool = model_summary_spool_mult(models)
print(spool)



    


















    






## Autoencoder ##
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         (None, 784)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 32)                25120     
_________________________________________________________________
dense_2 (Dense)              (None, 784)               25872     
=================================================================
Total params: 50,992
Trainable params: 50,992
Non-trainable params: 0
_________________________________________________________________
************************************************************

## Encoder ##
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_1 (InputLayer)         (None, 784)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 32)                25120     
=================================================================
Total params: 25,120
Trainable params: 25,120
Non-trainable params: 0
_________________________________________________________________
************************************************************

## Decoder ##
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         (None, 32)                0         
_________________________________________________________________
dense_2 (Dense)              (None, 784)               25872     
=================================================================
Total params: 25,872
Trainable params: 25,872
Non-trainable params: 0
_________________________________________________________________
************************************************************



















In [13]:



    


model_plot_save_mult(models)
model_display('autoencoder')



    


















    
























In [14]:



    


!ls -l



    


















    






total 504
-rw-r--r-- 1 root root   1426 Jun  7 19:18 autoencoder.json
-rw-r--r-- 1 root root   6849 Jun  7 19:18 autoencoder_plot.png
-rw-r--r-- 1 root root 218832 Jun  7 19:18 autoencoder_weights.h5
drwxr-xr-x 1 root root   4096 May 21 17:02 datalab
-rw-r--r-- 1 root root    896 Jun  7 19:18 decoder.json
-rw-r--r-- 1 root root   4380 Jun  7 19:18 decoder_plot.png
-rw-r--r-- 1 root root 114384 Jun  7 19:18 decoder_weights.h5
-rw-r--r-- 1 root root    893 Jun  7 19:18 encoder.json
-rw-r--r-- 1 root root   4028 Jun  7 19:18 encoder_plot.png
-rw-r--r-- 1 root root 111376 Jun  7 19:18 encoder_weights.h5
-rw-r--r-- 1 root root  21468 Jun  7 19:18 plot_predict_vs_reconstructed.png
-rw-r--r-- 1 root root   2159 Jun  7 19:18 spool_summaries.txt

















In [15]:



    


!zip -r data.zip ./ -x "datalab/*" "\.*"



    


















    






  adding: encoder_plot.png (deflated 7%)
  adding: encoder_weights.h5 (deflated 16%)
  adding: decoder_weights.h5 (deflated 15%)
  adding: autoencoder_plot.png (deflated 13%)
  adding: decoder_plot.png (deflated 7%)
  adding: plot_predict_vs_reconstructed.png (deflated 15%)
  adding: autoencoder_weights.h5 (deflated 13%)
  adding: spool_summaries.txt (deflated 89%)
  adding: decoder.json (deflated 58%)
  adding: encoder.json (deflated 58%)
  adding: autoencoder.json (deflated 72%)

















In [16]:



    


from google.colab import files
files.download('data.zip')



    











In [ ]:

Content source: xR86/ml-stuff

Similar notebooks:

notebook.community | gallery | about