def testTrainClassifierWithIntegerLabel(self):
def _input_fn_with_integer_label():
return (FEATURES_DICT,
constant_op.constant([[0], [1], [1], [0], [0]],
dtypes.int32))
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
est.train(input_fn=_input_fn_with_integer_label, steps=5)
self._assert_checkpoint(est.model_dir,
global_step=5,
finalized_trees=1,
attempted_layers=5)
eval_res = est.evaluate(input_fn=_input_fn_with_integer_label, steps=1)
self.assertAllClose(eval_res['accuracy'], 1.0)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose([[0], [1], [1], [0], [0]],
[pred['class_ids'] for pred in predictions])
def testInferBinaryClassifier(self):
train_input_fn = _make_train_input_fn(is_classification=True)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
# It will stop after 5 steps because of the max depth and num trees.
num_steps = 100
# Train for a few steps, and validate final checkpoint.
est.train(train_input_fn, steps=num_steps)
self._assert_checkpoint(est.model_dir,
global_step=5,
finalized_trees=1,
attempted_layers=5)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose([[0], [1], [1], [0], [0]],
[pred['class_ids'] for pred in predictions])
def testTrainClassifierWithLabelVocabulary(self):
apple, banana = 'apple', 'banana'
def _input_fn_with_label_vocab():
return FEATURES_DICT, [[apple], [banana], [banana], [apple],
[apple]]
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5,
label_vocabulary=[apple, banana])
est.train(input_fn=_input_fn_with_label_vocab, steps=5)
self._assert_checkpoint(est.model_dir,
global_step=5,
finalized_trees=1,
attempted_layers=5)
eval_res = est.evaluate(input_fn=_input_fn_with_label_vocab, steps=1)
self.assertAllClose(eval_res['accuracy'], 1.0)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose([[0], [1], [1], [0], [0]],
[pred['class_ids'] for pred in predictions])
def testInferBinaryClassifier(self):
train_input_fn = _make_train_input_fn(is_classification=True)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
# It will stop after 5 steps because of the max depth and num trees.
num_steps = 100
# Train for a few steps, and validate final checkpoint.
est.train(train_input_fn, steps=num_steps)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertEquals(5, len(predictions))
# All labels are correct.
self.assertAllClose([0], predictions[0]['class_ids'])
self.assertAllClose([1], predictions[1]['class_ids'])
self.assertAllClose([1], predictions[2]['class_ids'])
self.assertAllClose([0], predictions[3]['class_ids'])
self.assertAllClose([0], predictions[4]['class_ids'])
def testTrainAndEvaluateBinaryClassifier(self):
input_fn = _make_train_input_fn(is_classification=True)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
# It will stop after 5 steps because of the max depth and num trees.
num_steps = 100
# Train for a few steps, and validate final checkpoint.
est.train(input_fn, steps=num_steps)
self._assert_checkpoint(est.model_dir, 6)
eval_res = est.evaluate(input_fn=input_fn, steps=1)
self.assertAllClose(eval_res['accuracy'], 1.0)
def testTrainClassifierWithDataset(self):
train_input_fn = _make_train_input_fn_dataset(is_classification=True)
predict_input_fn = numpy_io.numpy_input_fn(
x=FEATURES_DICT, y=None, batch_size=1, num_epochs=1, shuffle=False)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
est.train(train_input_fn, steps=100) # will stop after 5 steps anyway.
self._assert_checkpoint(
est.model_dir, global_step=5, finalized_trees=1, attempted_layers=5)
eval_res = est.evaluate(input_fn=train_input_fn, steps=1)
self.assertAllClose(eval_res['accuracy'], 1.0)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose([[0], [1], [1], [0], [0]],
[pred['class_ids'] for pred in predictions])
def testTrainClassifierWithRankOneLabel(self):
"""Tests that label with rank-1 tensor is also accepted by classifier."""
def _input_fn_with_rank_one_label():
return FEATURES_DICT, [0., 1., 1., 0., 0.]
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
# It will stop after 5 steps because of the max depth and num trees.
num_steps = 100
# Train for a few steps, and validate final checkpoint.
est.train(_input_fn_with_rank_one_label, steps=num_steps)
self._assert_checkpoint(est.model_dir,
global_step=5,
finalized_trees=1,
attempted_layers=5)
eval_res = est.evaluate(input_fn=_input_fn_with_rank_one_label,
steps=1)
self.assertAllClose(eval_res['accuracy'], 1.0)
def testFirstCheckpointWorksFine(self):
"""Tests that eval/pred doesn't crash with the very first checkpoint.
The step-0 checkpoint will have only an empty ensemble, and a separate eval
job might read from it and crash.
This test ensures that prediction/evaluation works fine with it.
"""
input_fn = _make_train_input_fn(is_classification=True)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesClassifier(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
class BailOutWithoutTraining(session_run_hook.SessionRunHook):
def before_run(self, run_context):
raise StopIteration('to bail out.')
est.train(
input_fn,
steps=100, # must stop at 0 anyway.
hooks=[BailOutWithoutTraining()])
self._assert_checkpoint(est.model_dir,
global_step=0,
finalized_trees=0,
attempted_layers=0)
# Empty ensemble returns 0 logits, so that all output labels are 0.
eval_res = est.evaluate(input_fn=input_fn, steps=1)
self.assertAllClose(eval_res['accuracy'], 0.6)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose([[0], [0], [0], [0], [0]],
[pred['class_ids'] for pred in predictions])
