def testMultidimensionalAcculumator(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([]),
hessian_shape=tensor_shape.TensorShape([]))
with ops.control_dependencies([accumulator.initializer]):
op1 = accumulator.add(stamp_token=0,
partition_ids=[1, 2, 1],
feature_ids=[[2, 2], [3, 0], [2, 2]],
gradients=[0.1, 0.3, 0.8],
hessians=[0.2, 0.4, -9])
op2 = accumulator.add(0, [2, 1], [[3, 1], [2, 2]], [0.1, 1],
[0.2, -1])
with ops.control_dependencies([op1, op2]):
num_updates, partition, bucket_ids, grads, hessians = accumulator.flush(
stamp_token=0, next_stamp_token=1)
num_updates, partition, bucket_ids, grads, hessians = sess.run(
[num_updates, partition, bucket_ids, grads, hessians])
result = _AccumulatorResultToDict(partition, bucket_ids, grads,
hessians)
self.assertEqual(num_updates, 2)
self.assertEqual(len(result), 3)
# Key is partition, bucket, dimension.
self.assertAllClose(result[(1, 2, 2)], [1.9, -9.8])
self.assertAllClose(result[(2, 3, 0)], [0.3, 0.4])
self.assertAllClose(result[(2, 3, 1)], [0.1, 0.2])
def testDropStaleUpdate(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([]),
hessian_shape=tensor_shape.TensorShape([]))
with ops.control_dependencies([accumulator.initializer]):
op1 = accumulator.add(stamp_token=0,
partition_ids=[1, 2],
feature_ids=[[2, 0], [3, 0]],
gradients=[0.1, 0.3],
hessians=[0.2, 0.4])
op2 = accumulator.add(stamp_token=-1,
partition_ids=[1],
feature_ids=[[2, 0]],
gradients=[0.1],
hessians=[0.2])
with ops.control_dependencies([op1, op2]):
num_updates, partition, feature, grads, hessians = accumulator.flush(
stamp_token=0, next_stamp_token=1)
num_updates, partition, feature, grads, hessians = sess.run(
[num_updates, partition, feature, grads, hessians])
result = _AccumulatorResultToDict(partition, feature, grads,
hessians)
self.assertEqual(num_updates, 1)
self.assertEqual(len(result), 2)
self.assertAllClose(result[(1, 2, 0)], [0.1, 0.2])
self.assertAllClose(result[(2, 3, 0)], [0.3, 0.4])
def testDeserialize(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([]),
hessian_shape=tensor_shape.TensorShape([]))
with ops.control_dependencies([accumulator.initializer]):
# These will be deleted due to deserialize call.
op1 = accumulator.add(stamp_token=0,
partition_ids=[1, 2],
feature_ids=[[2, 0], [3, 1]],
gradients=[0.1, 0.3],
hessians=[0.2, 0.4])
with ops.control_dependencies([op1]):
deserialize = (accumulator.saveable.deserialize(
stamp_token=2,
num_updates=3,
partition_ids=[3, 4],
feature_ids=[[5, 0], [6, 2]],
gradients=[0.4, 0.5],
hessians=[0.6, 0.7]))
with ops.control_dependencies([deserialize]):
num_updates, partition, feature, grads, hessians = accumulator.flush(
stamp_token=2, next_stamp_token=3)
num_updates, partition, feature, grads, hessians = sess.run(
[num_updates, partition, feature, grads, hessians])
result = _AccumulatorResultToDict(partition, feature, grads,
hessians)
self.assertEqual(num_updates, 3)
self.assertEqual(len(result), 2)
self.assertAllClose(result[(3, 5, 0)], [0.4, 0.6])
self.assertAllClose(result[(4, 6, 2)], [0.5, 0.7])
def testMakeSummary(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([2]),
hessian_shape=tensor_shape.TensorShape([2, 2]))
partition, feature, grads, hessians = accumulator._make_summary(
partition_ids=[1, 2, 1],
feature_ids=[[2, 0], [3, 2], [2, 0]],
# Two values for gradients,
gradients=[[0.1, 0.1], [0.2, 0.2], [0.10, 0.11]],
# A 2x2 matrix for each hessian.
hessians=[[[0.01, 0.02], [0.03, 0.04]],
[[0.05, 0.06], [0.07, 0.08]],
[[0.011, 0.022], [0.033, 0.044]]])
partition, feature, grads, hessians = sess.run(
[partition, feature, grads, hessians])
result = _AccumulatorResultToDict(partition, feature, grads,
hessians)
self.assertEqual(len(result), 2)
self.assertAllClose(result[(1, 2, 0)][0], [0.20, 0.21])
self.assertAllClose(result[(1, 2, 0)][1],
[[0.021, 0.042], [0.063, 0.084]])
self.assertAllClose(result[(2, 3, 2)][0], [0.2, 0.2])
self.assertAllClose(result[(2, 3, 2)][1],
[[0.05, 0.06], [0.07, 0.08]])
def __init__(self,
l1_regularization,
l2_regularization,
tree_complexity_regularization,
min_node_weight,
feature_column_group_id,
epsilon,
num_quantiles,
gradient_shape,
hessian_shape,
multiclass_strategy,
init_stamp_token=0,
loss_uses_sum_reduction=False,
name=None):
"""Initialize the internal state for this split handler.
Args:
l1_regularization: L1 regularization applied for this split handler.
l2_regularization: L2 regularization applied for this split handler.
tree_complexity_regularization: Tree complexity regularization applied
for this split handler.
min_node_weight: Minimum sum of weights of examples in each partition to
be considered for splitting.
feature_column_group_id: Feature column group index.
epsilon: A float, the error bound for quantile computation.
num_quantiles: An int, the number of buckets to create from the histogram.
gradient_shape: A TensorShape, containing shape of gradients.
hessian_shape: A TensorShape, containing shape of hessians.
multiclass_strategy: Strategy describing how to treat multiclass problems.
init_stamp_token: A tensor containing an scalar for initial stamp of the
stamped objects.
loss_uses_sum_reduction: A scalar boolean tensor that specifies whether
SUM or MEAN reduction was used for the loss.
name: An optional handler name.
"""
super(InequalitySplitHandler, self).__init__(
name=name,
l1_regularization=l1_regularization,
l2_regularization=l2_regularization,
tree_complexity_regularization=tree_complexity_regularization,
min_node_weight=min_node_weight,
feature_column_group_id=feature_column_group_id,
gradient_shape=gradient_shape,
hessian_shape=hessian_shape,
multiclass_strategy=multiclass_strategy,
loss_uses_sum_reduction=loss_uses_sum_reduction)
self._stats_accumulator = stats_accumulator_ops.StatsAccumulator(
init_stamp_token,
gradient_shape,
hessian_shape,
name="StatsAccumulator/{}".format(self._name))
# Allocate both stats accumulator and quantile accumulator on the same
# device so that we can build splits with fewer RPCs.
with ops.colocate_with(self._stats_accumulator.resource_handle):
self._quantile_accumulator = quantile_ops.QuantileAccumulator(
init_stamp_token,
epsilon=epsilon,
num_quantiles=num_quantiles,
name="QuantileAccumulator/{}".format(self._name))
def __init__(
self,
sparse_int_column,
l1_regularization,
l2_regularization,
tree_complexity_regularization,
min_node_weight,
feature_column_group_id,
gradient_shape,
hessian_shape,
multiclass_strategy,
init_stamp_token=0,
loss_uses_sum_reduction=False,
weak_learner_type=learner_pb2.LearnerConfig.NORMAL_DECISION_TREE,
name=None):
"""Initialize the internal state for this split handler.
Args:
sparse_int_column: A `SparseTensor` column with int64 values associated
with this handler.
l1_regularization: L1 regularization applied for this split handler.
l2_regularization: L2 regularization applied for this split handler.
tree_complexity_regularization: Tree complexity regularization applied
for this split handler.
min_node_weight: Minimum sum of weights of examples in each partition to
be considered for splitting.
feature_column_group_id: Feature column group index.
gradient_shape: A TensorShape, containing shape of gradients.
hessian_shape: A TensorShape, containing shape of hessians.
multiclass_strategy: Strategy describing how to treat multiclass problems.
init_stamp_token: A tensor containing an scalar for initial stamp of the
stamped objects.
loss_uses_sum_reduction: A scalar boolean tensor that specifies whether
SUM or MEAN reduction was used for the loss.
weak_learner_type: Specifies the type of weak learner to use.
name: An optional handler name.
"""
super(EqualitySplitHandler, self).__init__(
l1_regularization=l1_regularization,
l2_regularization=l2_regularization,
tree_complexity_regularization=tree_complexity_regularization,
min_node_weight=min_node_weight,
feature_column_group_id=feature_column_group_id,
gradient_shape=gradient_shape,
hessian_shape=hessian_shape,
multiclass_strategy=multiclass_strategy,
loss_uses_sum_reduction=loss_uses_sum_reduction,
name=name)
self._stats_accumulator = stats_accumulator_ops.StatsAccumulator(
init_stamp_token,
gradient_shape,
hessian_shape,
name="StatsAccumulator/{}".format(self._name))
self._sparse_int_column = sparse_int_column
self._weak_learner_type = weak_learner_type
def testSerialize(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([2]),
hessian_shape=tensor_shape.TensorShape([2, 2]))
with ops.control_dependencies([accumulator.initializer]):
op1 = accumulator.add(
stamp_token=0,
partition_ids=[1, 2],
feature_ids=[[2, 0], [3, 0]],
# Two values for gradients,
gradients=[[0.1, 0.1], [0.2, 0.2]],
# A 2x2 matrix for each hessian.
hessians=[[[0.01, 0.02], [0.03, 0.04]],
[[0.05, 0.06], [0.07, 0.08]]])
with ops.control_dependencies([op1]):
(stamp_token, num_updates_1, partition_1, feature_1, grads_1,
hessians_1) = accumulator.saveable.serialize()
# Make sure that the accumulator hasn't changed during serialization.
with ops.control_dependencies([stamp_token]):
num_updates_2, partition_2, feature_2, grads_2, hessians_2 = (
accumulator.flush(stamp_token=0, next_stamp_token=1))
(stamp_token, num_updates_1, partition_1, feature_1, grads_1,
hessians_1, num_updates_2, partition_2, feature_2, grads_2,
hessians_2) = sess.run([
stamp_token, num_updates_1, partition_1, feature_1,
grads_1, hessians_1, num_updates_2, partition_2,
feature_2, grads_2, hessians_2
])
result_1 = _AccumulatorResultToDict(partition_1, feature_1,
grads_1, hessians_1)
result_2 = _AccumulatorResultToDict(partition_2, feature_2,
grads_2, hessians_2)
self.assertEqual(num_updates_1, 1)
self.assertEqual(num_updates_2, 1)
self.assertEqual(len(result_1), 2)
self.assertAllClose(result_1[(1, 2, 0)][0], [0.1, 0.1])
self.assertAllClose(result_1[(1, 2, 0)][1],
[[0.01, 0.02], [0.03, 0.04]])
self.assertAllClose(result_1[(2, 3, 0)][0], [0.2, 0.2])
self.assertAllClose(result_1[(2, 3, 0)][1],
[[0.05, 0.06], [0.07, 0.08]])
self.assertAllEqual(result_1[1, 2, 0][0], result_2[1, 2, 0][0])
self.assertAllEqual(result_1[1, 2, 0][1], result_2[1, 2, 0][1])
self.assertAllEqual(result_1[2, 3, 0][0], result_2[2, 3, 0][0])
self.assertAllEqual(result_1[2, 3, 0][1], result_2[2, 3, 0][1])
def testDeserialize(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([2]),
hessian_shape=tensor_shape.TensorShape([2, 2]))
with ops.control_dependencies([accumulator.initializer]):
# These will be deleted due to deserialize call.
op1 = accumulator.add(
stamp_token=0,
partition_ids=[1, 2],
feature_ids=[[2, 0], [3, 0]],
# Two values for gradients,
gradients=[[0.1, 0.1], [0.2, 0.2]],
# A 2x2 matrix for each hessian.
hessians=[[[0.01, 0.02], [0.03, 0.04]],
[[0.05, 0.06], [0.07, 0.08]]])
with ops.control_dependencies([op1]):
deserialize = accumulator.saveable.deserialize(
stamp_token=2,
num_updates=3,
partition_ids=[3, 4],
feature_ids=[[4, 0], [5, 0]],
# Two values for gradients,
gradients=[[0.3, 0.3], [0.5, 0.5]],
# A 2x2 matrix for each hessian.
hessians=[[[0.03, 0.04], [0.05, 0.06]],
[[0.07, 0.08], [0.09, 0.10]]])
with ops.control_dependencies([deserialize]):
num_updates, partition, feature, grads, hessians = accumulator.flush(
stamp_token=2, next_stamp_token=3)
num_updates, partition, feature, grads, hessians = sess.run(
[num_updates, partition, feature, grads, hessians])
result = _AccumulatorResultToDict(partition, feature, grads,
hessians)
self.assertEqual(num_updates, 3)
self.assertEqual(len(result), 2)
self.assertAllClose(result[(3, 4, 0)][0], [0.3, 0.3])
self.assertAllClose(result[(3, 4, 0)][1],
[[0.03, 0.04], [0.05, 0.06]])
self.assertAllClose(result[(4, 5, 0)][0], [0.5, 0.5])
self.assertAllClose(result[(4, 5, 0)][1],
[[0.07, 0.08], [0.09, 0.10]])
def testMakeSummary(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([]),
hessian_shape=tensor_shape.TensorShape([]))
partition, feature, grads, hessians = accumulator._make_summary(
partition_ids=[1, 2, 1],
feature_ids=[[2, 0], [3, 1], [2, 0]],
gradients=[0.1, 0.3, 0.1],
hessians=[0.2, 0.4, 0.2])
partition, feature, grads, hessians = sess.run(
[partition, feature, grads, hessians])
result = _AccumulatorResultToDict(partition, feature, grads,
hessians)
self.assertEqual(len(result), 2)
self.assertAllClose(result[(1, 2, 0)], [0.2, 0.4])
self.assertAllClose(result[(2, 3, 1)], [0.3, 0.4])
def testDropStaleUpdate(self):
with self.cached_session() as sess:
accumulator = stats_accumulator_ops.StatsAccumulator(
stamp_token=0,
gradient_shape=tensor_shape.TensorShape([2]),
hessian_shape=tensor_shape.TensorShape([2, 2]))
with ops.control_dependencies([accumulator.initializer]):
op1 = accumulator.add(
stamp_token=0,
partition_ids=[1, 2],
feature_ids=[[2, 5], [3, 0]],
# Two values for gradients,
gradients=[[0.1, 0.1], [0.2, 0.2]],
# A 2x2 matrix for each hessian.
hessians=[[[0.01, 0.02], [0.03, 0.04]],
[[0.05, 0.06], [0.07, 0.08]]])
op2 = accumulator.add(stamp_token=-1,
partition_ids=[1],
feature_ids=[[2, 5]],
gradients=[[0.10, 0.11]],
hessians=[[[0.011, 0.022],
[0.033, 0.044]]])
with ops.control_dependencies([op1, op2]):
num_updates, partition, feature, grads, hessians = accumulator.flush(
stamp_token=0, next_stamp_token=1)
num_updates, partition, feature, grads, hessians = sess.run(
[num_updates, partition, feature, grads, hessians])
result = _AccumulatorResultToDict(partition, feature, grads,
hessians)
self.assertEqual(num_updates, 1)
self.assertEqual(len(result), 2)
self.assertAllClose(result[(1, 2, 5)][0], [0.1, 0.1])
self.assertAllClose(result[(1, 2, 5)][1],
[[0.01, 0.02], [0.03, 0.04]])
self.assertAllClose(result[(2, 3, 0)][0], [0.2, 0.2])
self.assertAllClose(result[(2, 3, 0)][1],
[[0.05, 0.06], [0.07, 0.08]])