03_nn.py

# First Neural Net
# Train, evaluate, and predict with the model
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

import tensorflow as tf
from tensorflow import keras
import numpy as np
import matplotlib.pyplot as plt

mnist = keras.datasets.mnist

(x_train, y_train), (x_test, y_test) = mnist.load_data()
print(x_train.shape, y_train.shape)

# normalize: 0,255 -> 0,1
x_train, x_test = x_train / 255.0, x_test / 255.0

# model
model = keras.models.Sequential([
    keras.layers.Flatten(input_shape=(28,28)),
    keras.layers.Dense(128, activation='relu'),
    keras.layers.Dense(10),
])

print(model.summary())

# another way to build the Sequential model:
#model = keras.models.Sequential()
#model.add(keras.layers.Flatten(input_shape=(28,28))
#model.add(keras.layers.Dense(128, activation='relu'))
#model.add(keras.layers.Dense(10))

# loss and optimizer
loss = keras.losses.SparseCategoricalCrossentropy(from_logits=True)
optim = keras.optimizers.Adam(lr=0.001)
metrics = ["accuracy"]

model.compile(loss=loss, optimizer=optim, metrics=metrics)

# training
batch_size = 64
epochs = 5

model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, shuffle=True, verbose=2)

# evaulate
model.evaluate(x_test, y_test, batch_size=batch_size, verbose=2)

# predictions

# 1. option: build new model with Softmax layer
probability_model = keras.models.Sequential([
    model,
    keras.layers.Softmax()
])

predictions = probability_model(x_test)
pred0 = predictions[0]
print(pred0)

# use np.argmax to get label with highest probability
label0 = np.argmax(pred0)
print(label0)

# 2. option: original model + nn.softmax, call model(x)
predictions = model(x_test)
predictions = tf.nn.softmax(predictions)
pred0 = predictions[0]
print(pred0)
label0 = np.argmax(pred0)
print(label0)

# 3. option: original model + nn.softmax, call model.predict(x)
predictions = model.predict(x_test, batch_size=batch_size)
predictions = tf.nn.softmax(predictions)
pred0 = predictions[0]
print(pred0)
label0 = np.argmax(pred0)
print(label0)

# call argmax for multiple labels
pred05s = predictions[0:5]
print(pred05s.shape)
label05s = np.argmax(pred05s, axis=1)
print(label05s)