pdfcoffee

Chapter 3
4. Use the feature_column module of TensorFlow to define numeric features
of size 28×28:
feature_columns = [tf.feature_column.numeric_column("x",
shape=[28, 28])]
5. Create the logistic regression estimator. We use a simple LinearClassifier.
We encourage you to experiment with DNNClassifier as well:
classifier = tf.estimator.LinearClassifier(
feature_columns=feature_columns,
n_classes=10,
model_dir="mnist_model/"
)
6. Let us also build an input_function to feed the estimator:
train_input_fn = tf.compat.v1.estimator.inputs.numpy_input_fn(
x={"x": train_data},
y=train_labels,
batch_size=100,
num_epochs=None,
shuffle=True)
7. Let's now train the classifier:
classifier.train(input_fn=train_input_fn, steps=10)
8. Next, we create the input function for validation data:
val_input_fn = tf.compat.v1.estimator.inputs.numpy_input_fn(
x={"x": eval_data},
y=eval_labels,
num_epochs=1,
shuffle=False)
9. Let us evaluate the trained Linear Classifier on the validation data:
eval_results = classifier.evaluate(input_fn=val_input_fn)
print(eval_results)
10. We get an accuracy of 89.4% after 130 time steps. Not bad, right? Please note
that since we have specified the time steps, the model trains for the specified
steps and logs the value after 10 steps (the number of steps specified). Now
if we run train again, then it will start from the state it had at the 10th time
step. The steps will go on increasing with an increment of the number of
steps mentioned.
[ 105 ]