added autoencoder

keras-perceptron/autoencoder.py

@@ -0,0 +1,69 @@
+import numpy
+from keras.datasets import mnist
+from keras.models import Sequential
+from keras.layers import Dense, Flatten, Dropout, Reshape, ZeroPadding2D 
+from keras.layers import Conv2D, UpSampling2D, MaxPooling2D
+from keras.utils import np_utils
+import wandb
+from wandb.wandb_keras import WandbKerasCallback
+import random
+import cv2
+
+# logging code
+run = wandb.init()
+config = run.config
+
+# load data
+(X_train, _), (X_test, _) = mnist.load_data()
+
+img_width = X_train.shape[1]
+img_height = X_train.shape[2]
+
+
+X_train = X_train.astype('float32')
+X_train /= 255.
+X_test = X_test.astype('float32')
+X_test /= 255.
+
+X_train_noise = X_train + (numpy.random.randn(28*28*60000) * 0.2).reshape(60000, 28, 28) 
+X_train_noise.clip(0., 1.)
+
+X_test_noise = X_test + (numpy.random.randn(28*28*10000) * 0.2).reshape(10000, 28, 28) 
+X_test_noise.clip(0., 1.)
+
+# create model
+model=Sequential()
+#model.add(Flatten(input_shape=(img_width,img_height)))
+model.add(Reshape((28,28,1), input_shape=(28,28)))
+model.add(ZeroPadding2D((1,1)))
+model.add(Conv2D(32,
+    (3,3),
+    activation='relu'))
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(ZeroPadding2D((1,1), input_shape=(28, 28,1)))
+model.add(Conv2D(32,
+    (3,3),
+    activation='relu'))
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(UpSampling2D((2,2)))
+model.add(ZeroPadding2D((1,1), input_shape=(28, 28,1)))
+model.add(Conv2D(1,
+    (3,3),
+    activation='relu'))
+model.add(UpSampling2D((2,2)))
+model.add(Reshape((28,28)))
+model.summary()
+
+model.compile(loss='mse', optimizer='adam',
+                metrics=['accuracy'])
+
+# Fit the model
+model.fit(X_train_noise, X_train, epochs=10, validation_data=(X_test_noise, X_test),
+                    callbacks=[WandbKerasCallback()])
+
+X_pred = model.predict([X_test_noise[:10]])
+print(X_pred)
+X_test_noise *= 255
+cv2.imwrite('input.png', X_test_noise[0].reshape(28,28,1))
+X_pred *= 255.
+cv2.imwrite('output.png', X_pred[0].reshape(28,28,1))