def testBasicQuantileBuckets(self):
"""Sets up the quantile summary op test as follows.
Create a batch of 6 examples having a dense and sparse features.
The data looks like this
| Instance | instance weights | Dense 0 | Sparse 0
| 0 | 10 | 1 |
| 1 | 1 | 2 | 2
| 2 | 1 | 3 | 3
| 3 | 1 | 4 | 4
| 4 | 1 | 4 | 5
| 5 | 1 | 5 | 6
"""
dense_float_tensor_0 = constant_op.constant(
[1, 2, 3, 4, 4, 5], dtype=dtypes.float32)
sparse_indices_0 = constant_op.constant(
[[1, 0], [2, 0], [3, 0], [4, 0], [5, 0]], dtype=dtypes.int64)
sparse_values_0 = constant_op.constant(
[2, 3, 4, 5, 6], dtype=dtypes.float32)
sparse_shape_0 = constant_op.constant([6, 1], dtype=dtypes.int64)
example_weights = constant_op.constant(
[10, 1, 1, 1, 1, 1], dtype=dtypes.float32)
with self.test_session():
config = self._gen_config(0.33, 3)
dense_buckets, sparse_buckets = quantile_ops.quantile_buckets(
[dense_float_tensor_0], [sparse_indices_0], [sparse_values_0],
[sparse_shape_0],
example_weights=example_weights,
dense_config=[config],
sparse_config=[config])
self.assertAllEqual([1, 3, 5], dense_buckets[0].eval())
self.assertAllEqual([2, 4, 6.], sparse_buckets[0].eval())
def testFixedNonUniform(self):
"""Sets up the quantile summary op test as follows.
Creates array dividing range [0, 1] to 1<<16 elements equally spaced
with weight same as the value.
"""
dense_float_tensor_0 = constant_op.constant(
[(1.0 * i) / math.pow(2.0, 16)
for i in range(0, int(math.pow(2, 16)) + 1)])
example_weights = constant_op.constant(
[(1.0 * i) / math.pow(2.0, 16)
for i in range(0, int(math.pow(2, 16)) + 1)])
config = self._gen_config(0.1, 10)
with self.test_session():
dense_buckets, _ = quantile_ops.quantile_buckets(
[dense_float_tensor_0], [], [], [],
example_weights=example_weights,
dense_config=[config],
sparse_config=[])
self.assertAllClose(
[0] + [math.sqrt((i + 1.0) / 10) for i in range(0, 10)],
dense_buckets[0].eval(),
atol=0.1)
def testBasicQuantileBuckets(self):
"""Sets up the quantile summary op test as follows.
Create a batch of 6 examples having a dense and sparse features. SparseM is
a sparse multi-dimensional (multivalent) feature.
The data looks like this
| Instance | instance weights | Dense 0 | Sparse 0 | SparseM
| 0 | 10 | 1 | | | |
| 1 | 1 | 2 | 2 | 2 | |
| 2 | 1 | 3 | 3 | 3 | |
| 3 | 1 | 4 | 4 | | 4 |
| 4 | 1 | 4 | 5 | | 5 |
| 5 | 1 | 5 | 6 | | 6 |
"""
dense_float_tensor_0 = constant_op.constant([1, 2, 3, 4, 4, 5],
dtype=dtypes.float32)
sparse_indices_0 = constant_op.constant(
[[1, 0], [2, 0], [3, 0], [4, 0], [5, 0]], dtype=dtypes.int64)
sparse_values_0 = constant_op.constant([2, 3, 4, 5, 6],
dtype=dtypes.float32)
sparse_shape_0 = constant_op.constant([6, 1], dtype=dtypes.int64)
# Multi-dimensional feature that should have the same quantiles as Sparse 0.
sparse_indices_m = constant_op.constant(
[[1, 1], [2, 0], [3, 1], [4, 1], [5, 1]], dtype=dtypes.int64)
sparse_values_m = constant_op.constant([2, 3, 4, 5, 6],
dtype=dtypes.float32)
sparse_shape_m = constant_op.constant([6, 2], dtype=dtypes.int64)
example_weights = constant_op.constant([10, 1, 1, 1, 1, 1],
dtype=dtypes.float32)
with self.test_session():
config = self._gen_config(0.33, 3)
dense_buckets, sparse_buckets = quantile_ops.quantile_buckets(
[dense_float_tensor_0], [sparse_indices_0, sparse_indices_m],
[sparse_values_0, sparse_values_m],
[sparse_shape_0, sparse_shape_m],
example_weights=example_weights,
dense_config=[config],
sparse_config=[config, config])
self.assertAllEqual([1, 3, 5], dense_buckets[0].eval())
self.assertAllEqual([2, 4, 6.], sparse_buckets[0].eval())
# Multidimensional sparse.
self.assertAllEqual([2, 4, 6.], sparse_buckets[1].eval())
def testBasicQuantileBuckets(self):
"""Sets up the quantile summary op test as follows.
Create a batch of 6 examples having a dense and sparse features. SparseM is
a sparse multi-dimensional (multivalent) feature.
The data looks like this
| Instance | instance weights | Dense 0 | Sparse 0 | SparseM
| 0 | 10 | 1 | | | |
| 1 | 1 | 2 | 2 | 2 | |
| 2 | 1 | 3 | 3 | 3 | |
| 3 | 1 | 4 | 4 | | 4 |
| 4 | 1 | 4 | 5 | | 5 |
| 5 | 1 | 5 | 6 | | 6 |
"""
dense_float_tensor_0 = constant_op.constant(
[1, 2, 3, 4, 4, 5], dtype=dtypes.float32)
sparse_indices_0 = constant_op.constant(
[[1, 0], [2, 0], [3, 0], [4, 0], [5, 0]], dtype=dtypes.int64)
sparse_values_0 = constant_op.constant(
[2, 3, 4, 5, 6], dtype=dtypes.float32)
sparse_shape_0 = constant_op.constant([6, 1], dtype=dtypes.int64)
# Multi-dimensional feature that should have the same quantiles as Sparse 0.
sparse_indices_m = constant_op.constant(
[[1, 1], [2, 0], [3, 1], [4, 1], [5, 1]], dtype=dtypes.int64)
sparse_values_m = constant_op.constant(
[2, 3, 4, 5, 6], dtype=dtypes.float32)
sparse_shape_m = constant_op.constant([6, 2], dtype=dtypes.int64)
example_weights = constant_op.constant(
[10, 1, 1, 1, 1, 1], dtype=dtypes.float32)
with self.test_session():
config = self._gen_config(0.33, 3)
dense_buckets, sparse_buckets = quantile_ops.quantile_buckets(
[dense_float_tensor_0], [sparse_indices_0, sparse_indices_m],
[sparse_values_0, sparse_values_m], [sparse_shape_0, sparse_shape_m],
example_weights=example_weights,
dense_config=[config],
sparse_config=[config, config])
self.assertAllEqual([1, 3, 5], dense_buckets[0].eval())
self.assertAllEqual([2, 4, 6.], sparse_buckets[0].eval())
# Multidimensional sparse.
self.assertAllEqual([2, 4, 6.], sparse_buckets[1].eval())
