training and scoring

Author: lutsker

README.md

@@ -0,0 +1 @@
+MNIST handwritten digits with a convolutional neural network. 

data/__init__.py

@@ -0,0 +1 @@
+from .data import dataset 

data/data.py

@@ -0,0 +1,15 @@
+import numpy as np 
+from tensorflow.keras import datasets 
+from tensorflow.keras.utils import to_categorical
+
+def dataset():
+    (x_train, y_train), (x_test, y_test) = datasets.mnist.load_data()
+    x_train = x_train.astype("float32") / 255
+    x_test = x_test.astype("float32") / 255
+    x_train = np.expand_dims(x_train, -1)
+    x_test = np.expand_dims(x_test, -1)
+    y_train = to_categorical(y_train, 10)
+    y_test = to_categorical(y_test, 10)    
+    def result(set):
+        return (x_train, y_train) if set == 'train' else (x_test, y_test)
+    return result 

main.py

@@ -1,4 +1,15 @@
 from model import model_factory
+from data import dataset
+from trainer import trainer
 
 model = model_factory()
-model.summary()
+print(model.summary())
+data = dataset()
+
+train = trainer(model, data)
+model = train(epochs=10, batch_size=128)
+
+(x, y) = data('test')
+score = model.evaluate(x, y, verbose=0)
+print("Test loss:", score[0])
+print("Test accuracy:", score[1])

model/layers.py

@@ -10,7 +10,7 @@
 def Conv(filters, kernel_size, block_name):
     def result(input):
         conv = Conv2D(filters, kernel_size, padding="same", name=block_name)(input)
-        relu  = Activation('relu')(conv)
+        relu = Activation('relu')(conv)
         pool = MaxPooling2D(pool_size=(2, 2))(relu)
         return pool
     return result
@@ -20,8 +20,8 @@ def result(input):
         flatten = Flatten(name='flat')(input)
         dropout = Dropout(0.5, name='drop')(flatten)
         dense = Dense(units=num_classes)(dropout)
-        relu = Activation('relu', name='relu_x')(dense)
-        return relu
+        softmax = Activation('softmax', name='softmax_x')(dense)
+        return softmax
     return result
 
 def Inputs():

trainer/__init__.py

@@ -0,0 +1 @@
+from .trainer import trainer

trainer/trainer.py

@@ -0,0 +1,10 @@
+
+
+def trainer(model, data):
+    model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
+    (x, y) = data('train')
+    def result(epochs, batch_size):
+        model.fit(x, y, batch_size=batch_size, epochs=epochs, validation_split=0.1)
+        return model 
+    return result
+