1
model_dir = "no_regulation"
print(model_dir)
m = tf.contrib.learn.LinearClassifier(
feature_columns=feature_columns,
optimizer=tf.train.FtrlOptimizer(
learning_rate=3,
l1_regularization_strength=0,
l2_regularization_strength=0),
n_classes = n_classes,
model_dir=model_dir)
def train_input_fn():
print("Here!")
return input_fn(train.sample(50000), label_column = "course_index", categorical_columns = CATEGORICAL_COLUMNS)
,如果我这样做时,它批处理50000个样品每10步,如何在tf.learn中使用input_fn进行批量训练?
for i in range(40):
for j in range(20):
m.fit(input_fn=train_input_fn, steps = 10)
m.evaluate(input_fn=eval_input_fn1, steps = 1, name="test1")
m.evaluate(input_fn=eval_input_fn2, steps = 1, name="test2")
,这是否合理?如果我做m.fit(input_fn = train_input_fn,步= 1),每一个适合的通话将创建检查点,并会减慢培养了不少。我应该禁用检查点?如果是这样,怎么样?