Keras != tf.keras

• Keras是一个框架

• datasets

• layers

• losses

• metrics

• optimizers

Outline1

• Metrics

• update_state

• result().numpy()

• reset_states

Metrics

Step1.Build a meter

acc_meter = metrics.Accuarcy()
loss_meter = metrics.Mean

Step2.Update data

loss_meter.update_state(loss)
acc_meter.update_state(y,pred)

Step3.Get Average data

print(step, ‘loss:‘, loss_meter.result().numpy())
# ...
print(step,‘Evaluate Acc:‘, total_correct/total, acc_meter.result().numpy()

Clear buffer

if step % 100 == 0:
    print(step, ‘loss:‘, loss_meter.result().numpy())
    loss_meter.reset_states()
# ...
if step % 500 == 0:

total, total_correct = 0., 0

acc_meter.reset_states()

Outline2

• Compile

• Fit

• Evaluate

• Predict

Compile + Fit

Individual loss and optimize1

with tf.GradientTape() as tape:
    x = tf.reshape(x, (-1, 28*28))
    out = network(x)
    y_onehot = tf.one_hot(y, depth=10)
    loss = tf.reduce_mean(tf.losses.categorical_crossentropy(y_onehot, out, from_logits=True))
grads = tape.gradient(loss, network.trainable_variables)

optimizer.apply_gradients(zip(grads, network.trainable_variables))

Now1

network.compile(optimizer=optimizers.Adam(lr=0.01),
                loss=tf.losses.CategoricalCrossentropy(fromlogits=True),
                metircs=[‘accuracy‘])

Individual epoch and step2

for epoch in range(epochs):
    for step, (x, y) in enumerate(db):
        # ...

Now2

network.compile(optimizer=optimizers.Adam(lr=0.01),
                loss=tf.losses.CategoricalCrossentropy(fromlogits=True),
                metircs=[‘accuracy‘])
network.fit(db, epochs=10)

Standard Progressbar

Individual evaluation3

if step % 500 == 0:
    total, total_correct = 0., 0
<span class="hljs-keyword">for</span> step, (x, y) <span class="hljs-keyword">in</span> <span class="hljs-built_in">enumerate</span>(ds_val):
    x = tf.reshape(x, (<span class="hljs-number">-1</span>, <span class="hljs-number">28</span>*<span class="hljs-number">28</span>))
    out = network(x)
    pred = tf.argmax(out, axis=<span class="hljs-number">1</span>)
    pred = tf.cast(pred, dtype=tf.int32)
    correct = tf.equal(pred, y)
    total_correct += tf.reduce_sum(tf.cast(correct, dtype=tf.int32)).numpy()
    total += x.shape[<span class="hljs-number">0</span>]
   
print(step, <span class="hljs-string">‘Evaluate Acc:‘</span>, total_correct/total)

Now3

network.compile(optimizer=optimizers.Adam(lr=0.01),
                loss=tf.losses.CategoricalCrossentropy(fromlogits=True),
                metircs=[‘accuracy‘])
# validation_freq=2表示2个epochs做一次验证

network.fit(db, epochs=10, validation_data=ds_val, validation_freq=2)

Evaluation

Test

network.compile(optimizer=optimizers.Adam(lr=0.01),
                loss=tf.losses.CategoricalCrossentropy(fromlogits=True),
                metircs=[‘accuracy‘])
# validation_freq=2表示2个epochs做一次验证

network.fit(db, epochs=10, validation_data=ds_val, validation_freq=2)
network.evaluate(ds_val)

Predict

sample = next(iter(ds_val))
x = sample[0]
y = sample[1]
pred = network.predict(x)
y = tf.argmax(y, axis=1)
pred = tf.argmax(pre, axis=1)
print(pred)

print(y)