def testTrainRegressorWithDatasetSmallerBatch(self):
# Even when using small batches, if (n_batches_per_layer * batch_size) makes
# the same entire data size, the result should be the same.
train_input_fn = _make_train_input_fn_dataset(is_classification=False,
batch=1)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesRegressor(
feature_columns=self._feature_columns,
n_batches_per_layer=5,
n_trees=1,
max_depth=5)
# Train stops after (n_batches_per_layer * n_trees * max_depth) steps.
est.train(train_input_fn, steps=100)
self._assert_checkpoint(est.model_dir,
global_step=25,
finalized_trees=1,
attempted_layers=5)
# 5 batches = one epoch.
eval_res = est.evaluate(input_fn=train_input_fn, steps=5)
self.assertAllClose(eval_res['average_loss'], 2.478283)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose(
[[0.571619], [0.262821], [0.124549], [0.956801], [1.769801]],
[pred['predictions'] for pred in predictions])
def testTrainRegressorWithDatasetWhenInputIsOverEarlier(self):
train_input_fn = _make_train_input_fn_dataset(
is_classification=False, repeat=3) # to stop input after 3 steps.
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesRegressor(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=1,
max_depth=5)
# Note that training will stop when input exhausts.
# This might not be a typical pattern, but dataset.repeat(3) causes
# the input stream to cease after 3 steps.
est.train(train_input_fn, steps=100)
self._assert_checkpoint(est.model_dir,
global_step=3,
finalized_trees=0,
attempted_layers=3)
eval_res = est.evaluate(input_fn=train_input_fn, steps=1)
self.assertAllClose(eval_res['average_loss'], 3.777295)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose(
[[0.353850], [0.254100], [0.106850], [0.712100], [1.012100]],
[pred['predictions'] for pred in predictions])
def testTrainRegressorWithDatasetLargerBatch(self):
# The batch_size as the multiple of the entire data size should still yield
# the same result.
train_input_fn = _make_train_input_fn_dataset(is_classification=False,
batch=15)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesRegressor(
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['average_loss'], 2.478283)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertAllClose(
[[0.571619], [0.262821], [0.124549], [0.956801], [1.769801]],
[pred['predictions'] for pred in predictions])
def testInferRegressor(self):
train_input_fn = _make_train_input_fn(is_classification=False)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesRegressor(
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(
[[0.571619], [0.262821], [0.124549], [0.956801], [1.769801]],
[pred['predictions'] for pred in predictions])
def testInferRegressor(self):
train_input_fn = _make_train_input_fn(is_classification=False)
predict_input_fn = numpy_io.numpy_input_fn(x=FEATURES_DICT,
y=None,
batch_size=1,
num_epochs=1,
shuffle=False)
est = boosted_trees.BoostedTreesRegressor(
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, 6)
predictions = list(est.predict(input_fn=predict_input_fn))
self.assertEquals(5, len(predictions))
self.assertAllClose([0.703549], predictions[0]['predictions'])
self.assertAllClose([0.266539], predictions[1]['predictions'])
self.assertAllClose([0.256479], predictions[2]['predictions'])
self.assertAllClose([1.088732], predictions[3]['predictions'])
self.assertAllClose([1.901732], predictions[4]['predictions'])
def testTrainEvaluateAndPredictWithIndicatorColumn(self):
categorical = feature_column.categorical_column_with_vocabulary_list(
key='categorical', vocabulary_list=('bad', 'good', 'ok'))
feature_indicator = feature_column.indicator_column(categorical)
bucketized_col = feature_column.bucketized_column(
feature_column.numeric_column('an_uninformative_feature',
dtype=dtypes.float32),
BUCKET_BOUNDARIES)
labels = np.array([[0.], [5.7], [5.7], [0.], [0.]], dtype=np.float32)
# Our categorical feature defines the labels perfectly
input_fn = numpy_io.numpy_input_fn(x={
'an_uninformative_feature':
np.array([1, 1, 1, 1, 1]),
'categorical':
np.array(['bad', 'good', 'good', 'ok', 'bad']),
},
y=labels,
batch_size=5,
shuffle=False)
# Train depth 1 tree.
est = boosted_trees.BoostedTreesRegressor(
feature_columns=[bucketized_col, feature_indicator],
n_batches_per_layer=1,
n_trees=1,
learning_rate=1.0,
max_depth=1)
num_steps = 1
est.train(input_fn, steps=num_steps)
ensemble = self._assert_checkpoint_and_return_model(est.model_dir,
global_step=1,
finalized_trees=1,
attempted_layers=1)
# We learnt perfectly.
eval_res = est.evaluate(input_fn=input_fn, steps=1)
self.assertAllClose(eval_res['loss'], 0)
predictions = list(est.predict(input_fn))
self.assertAllClose(labels,
[pred['predictions'] for pred in predictions])
self.assertEqual(3, len(ensemble.trees[0].nodes))
# Check that the split happened on 'good' value, which will be encoded as
# feature with index 2 (0-numeric, 1 - 'bad')
self.assertEqual(
2, ensemble.trees[0].nodes[0].bucketized_split.feature_id)
self.assertEqual(0,
ensemble.trees[0].nodes[0].bucketized_split.threshold)
def testTrainAndEvaluateRegressor(self):
input_fn = _make_train_input_fn(is_classification=False)
est = boosted_trees.BoostedTreesRegressor(
feature_columns=self._feature_columns,
n_batches_per_layer=1,
n_trees=2,
max_depth=5)
# It will stop after 10 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, 11)
eval_res = est.evaluate(input_fn=input_fn, steps=1)
self.assertAllClose(eval_res['average_loss'], 0.913176)
def testTrainRegressorWithRankOneLabel(self):
"""Tests that label with rank-1 tensor is also accepted by regressor."""
def _input_fn_with_rank_one_label():
return FEATURES_DICT, [1.5, 0.3, 0.2, 2., 5.]
est = boosted_trees.BoostedTreesRegressor(
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['average_loss'], 2.478283)
