Python tensorflow.python.ops.math_ops.to_double() Examples
The following are 30
code examples of tensorflow.python.ops.math_ops.to_double().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
tensorflow.python.ops.math_ops
, or try the search function
.
.
Example #1
| Source File: metric_ops.py From deep_image_model with Apache License 2.0 | 5 votes |
|
def _sparse_false_negative_at_k(predictions_idx,
labels,
class_id=None,
weights=None):
"""Calculates false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose shape is broadcastable to the the first [D1, ... DN]
dimensions of `predictions_idx` and `labels`.
Returns:
A [D1, ... DN] `Tensor` of false negative counts.
"""
with ops.name_scope(None, 'false_negatives', (predictions_idx, labels)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fn = set_ops.set_size(set_ops.set_difference(predictions_idx,
labels,
aminusb=False))
fn = math_ops.to_double(fn)
if weights is not None:
weights = math_ops.to_double(weights)
fn = math_ops.mul(fn, weights)
return fn
Example #2
| Source File: metric_ops.py From deep_image_model with Apache License 2.0 | 5 votes |
|
def _sparse_false_positive_at_k(predictions_idx,
labels,
class_id=None,
weights=None):
"""Calculates false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose shape is broadcastable to the the first [D1, ... DN]
dimensions of `predictions_idx` and `labels`.
Returns:
A [D1, ... DN] `Tensor` of false positive counts.
"""
with ops.name_scope(None, 'false_positives', (predictions_idx, labels)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fp = set_ops.set_size(set_ops.set_difference(
predictions_idx, labels, aminusb=True))
fp = math_ops.to_double(fp)
if weights is not None:
weights = math_ops.to_double(weights)
fp = math_ops.mul(fp, weights)
return fp
Example #3
| Source File: metric_ops.py From deep_image_model with Apache License 2.0 | 5 votes |
|
def _sparse_true_positive_at_k(predictions_idx,
labels,
class_id=None,
weights=None,
name=None):
"""Calculates true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose shape is broadcastable to the the first [D1, ... DN]
dimensions of `predictions_idx` and `labels`.
name: Name of operation.
Returns:
A [D1, ... DN] `Tensor` of true positive counts.
"""
with ops.name_scope(name, 'true_positives', (predictions_idx, labels)):
labels, predictions_idx = _maybe_select_class_id(
labels, predictions_idx, class_id)
tp = set_ops.set_size(set_ops.set_intersection(predictions_idx, labels))
tp = math_ops.to_double(tp)
if weights is not None:
weights = math_ops.to_double(weights)
tp = math_ops.mul(tp, weights)
return tp
Example #4
| Source File: metrics_impl.py From keras-lambda with MIT License | 4 votes |
|
def _streaming_sparse_true_positive_at_k(labels,
predictions_idx,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
with ops.name_scope(
name, _at_k_name('true_positive', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
tp = _sparse_true_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_tp = math_ops.to_double(math_ops.reduce_sum(tp))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_tp, name='update')
Example #5
| Source File: metrics_impl.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
|
def _streaming_sparse_true_positive_at_k(labels,
predictions_idx,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incompatible shape.
"""
with ops.name_scope(
name, _at_k_name('true_positive', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
tp = _sparse_true_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_tp = math_ops.to_double(math_ops.reduce_sum(tp))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_tp, name='update')
Example #6
| Source File: metrics_impl.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
|
def _sparse_false_negative_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false negative counts.
"""
with ops.name_scope(
None, 'false_negatives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fn = sets.set_size(sets.set_difference(predictions_idx,
labels,
aminusb=False))
fn = math_ops.to_double(fn)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fn),)):
weights = math_ops.to_double(weights)
fn = math_ops.multiply(fn, weights)
return fn
Example #7
| Source File: metrics_impl.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
|
def _streaming_sparse_false_negative_at_k(labels,
predictions_idx,
k,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incompatible shape.
"""
with ops.name_scope(
name, _at_k_name('false_negative', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
fn = _sparse_false_negative_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_fn = math_ops.to_double(math_ops.reduce_sum(fn))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_fn, name='update')
Example #8
| Source File: metrics_impl.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
|
def _sparse_false_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false positive counts.
"""
with ops.name_scope(
None, 'false_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fp = sets.set_size(sets.set_difference(
predictions_idx, labels, aminusb=True))
fp = math_ops.to_double(fp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fp),)):
weights = math_ops.to_double(weights)
fp = math_ops.multiply(fp, weights)
return fp
Example #9
| Source File: image_ops_impl.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
|
def central_crop(image, central_fraction):
"""Crop the central region of the image.
Remove the outer parts of an image but retain the central region of the image
along each dimension. If we specify central_fraction = 0.5, this function
returns the region marked with "X" in the below diagram.
--------
| |
| XXXX |
| XXXX |
| | where "X" is the central 50% of the image.
--------
Args:
image: 3-D float Tensor of shape [height, width, depth]
central_fraction: float (0, 1], fraction of size to crop
Raises:
ValueError: if central_crop_fraction is not within (0, 1].
Returns:
3-D float Tensor
"""
image = ops.convert_to_tensor(image, name='image')
if central_fraction <= 0.0 or central_fraction > 1.0:
raise ValueError('central_fraction must be within (0, 1]')
if central_fraction == 1.0:
return image
image = control_flow_ops.with_dependencies(
_Check3DImage(image, require_static=False), image)
img_shape = array_ops.shape(image)
depth = image.get_shape()[2]
img_h = math_ops.to_double(img_shape[0])
img_w = math_ops.to_double(img_shape[1])
bbox_h_start = math_ops.to_int32((img_h - img_h * central_fraction) / 2)
bbox_w_start = math_ops.to_int32((img_w - img_w * central_fraction) / 2)
bbox_h_size = img_shape[0] - bbox_h_start * 2
bbox_w_size = img_shape[1] - bbox_w_start * 2
bbox_begin = array_ops.stack([bbox_h_start, bbox_w_start, 0])
bbox_size = array_ops.stack([bbox_h_size, bbox_w_size, -1])
image = array_ops.slice(image, bbox_begin, bbox_size)
# The first two dimensions are dynamic and unknown.
image.set_shape([None, None, depth])
return image
Example #10
| Source File: metrics_impl.py From keras-lambda with MIT License | 4 votes |
|
def _sparse_true_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None,
name=None):
"""Calculates true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of operation.
Returns:
A [D1, ... DN] `Tensor` of true positive counts.
"""
with ops.name_scope(
name, 'true_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(
labels, predictions_idx, class_id)
tp = sets.set_size(sets.set_intersection(predictions_idx, labels))
tp = math_ops.to_double(tp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, tp),)):
weights = math_ops.to_double(weights)
tp = math_ops.multiply(tp, weights)
return tp
Example #11
| Source File: metric_ops.py From deep_image_model with Apache License 2.0 | 4 votes |
|
def _streaming_sparse_false_positive_at_k(predictions_idx,
labels,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose shape is broadcastable to the the first [D1, ... DN]
dimensions of `predictions_idx` and `labels`.
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
default_name = _at_k_name('false_positive', k, class_id=class_id)
with ops.name_scope(name, default_name, (predictions_idx, labels)) as scope:
fp = _sparse_false_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_fp = math_ops.to_double(math_ops.reduce_sum(fp))
var = contrib_variables.local_variable(
array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_fp, name='update')
Example #12
| Source File: metrics_impl.py From keras-lambda with MIT License | 4 votes |
|
def _sparse_false_negative_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false negative counts.
"""
with ops.name_scope(
None, 'false_negatives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fn = sets.set_size(sets.set_difference(predictions_idx,
labels,
aminusb=False))
fn = math_ops.to_double(fn)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fn),)):
weights = math_ops.to_double(weights)
fn = math_ops.multiply(fn, weights)
return fn
Example #13
| Source File: metrics_impl.py From keras-lambda with MIT License | 4 votes |
|
def _streaming_sparse_false_negative_at_k(labels,
predictions_idx,
k,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
with ops.name_scope(
name, _at_k_name('false_negative', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
fn = _sparse_false_negative_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_fn = math_ops.to_double(math_ops.reduce_sum(fn))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_fn, name='update')
Example #14
| Source File: metrics_impl.py From keras-lambda with MIT License | 4 votes |
|
def _sparse_false_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false positive counts.
"""
with ops.name_scope(
None, 'false_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fp = sets.set_size(sets.set_difference(
predictions_idx, labels, aminusb=True))
fp = math_ops.to_double(fp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fp),)):
weights = math_ops.to_double(weights)
fp = math_ops.multiply(fp, weights)
return fp
Example #15
| Source File: metric_ops.py From keras-lambda with MIT License | 4 votes |
|
def _sparse_true_positive_at_k(predictions_idx,
labels,
class_id=None,
weights=None,
name=None):
"""Calculates true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of operation.
Returns:
A [D1, ... DN] `Tensor` of true positive counts.
"""
with ops.name_scope(
name, 'true_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(
labels, predictions_idx, class_id)
tp = set_ops.set_size(set_ops.set_intersection(predictions_idx, labels))
tp = math_ops.to_double(tp)
if weights is not None:
weights = math_ops.to_double(weights)
with ops.control_dependencies((_assert_weights_rank(weights, tp),)):
tp = math_ops.multiply(tp, weights)
return tp
Example #16
| Source File: metric_ops.py From keras-lambda with MIT License | 4 votes |
|
def _streaming_sparse_true_positive_at_k(predictions_idx,
labels,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
default_name = _at_k_name('true_positive', k, class_id=class_id)
with ops.name_scope(
name, default_name, (predictions_idx, labels, weights)) as scope:
tp = _sparse_true_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_tp = math_ops.to_double(math_ops.reduce_sum(tp))
var = contrib_variables.local_variable(
array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_tp, name='update')
Example #17
| Source File: metric_ops.py From keras-lambda with MIT License | 4 votes |
|
def _sparse_false_positive_at_k(predictions_idx,
labels,
class_id=None,
weights=None):
"""Calculates false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false positive counts.
"""
with ops.name_scope(
None, 'false_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fp = set_ops.set_size(set_ops.set_difference(
predictions_idx, labels, aminusb=True))
fp = math_ops.to_double(fp)
if weights is not None:
weights = math_ops.to_double(weights)
with ops.control_dependencies((_assert_weights_rank(weights, fp),)):
fp = math_ops.multiply(fp, weights)
return fp
Example #18
| Source File: metric_ops.py From keras-lambda with MIT License | 4 votes |
|
def _streaming_sparse_false_positive_at_k(predictions_idx,
labels,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
with ops.name_scope(
name, _at_k_name('false_positive', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
fp = _sparse_false_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_fp = math_ops.to_double(math_ops.reduce_sum(fp))
var = contrib_variables.local_variable(
array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_fp, name='update')
Example #19
| Source File: metric_ops.py From keras-lambda with MIT License | 4 votes |
|
def _sparse_false_negative_at_k(predictions_idx,
labels,
class_id=None,
weights=None):
"""Calculates false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false negative counts.
"""
with ops.name_scope(
None, 'false_negatives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fn = set_ops.set_size(set_ops.set_difference(predictions_idx,
labels,
aminusb=False))
fn = math_ops.to_double(fn)
if weights is not None:
weights = math_ops.to_double(weights)
with ops.control_dependencies((_assert_weights_rank(weights, fn),)):
fn = math_ops.multiply(fn, weights)
return fn
Example #20
| Source File: official_tf_image.py From X-Detector with Apache License 2.0 | 4 votes |
|
def central_crop(image, central_fraction):
"""Crop the central region of the image.
Remove the outer parts of an image but retain the central region of the image
along each dimension. If we specify central_fraction = 0.5, this function
returns the region marked with "X" in the below diagram.
--------
| |
| XXXX |
| XXXX |
| | where "X" is the central 50% of the image.
--------
Args:
image: 3-D float Tensor of shape [height, width, depth]
central_fraction: float (0, 1], fraction of size to crop
Raises:
ValueError: if central_crop_fraction is not within (0, 1].
Returns:
3-D float Tensor
"""
image = ops.convert_to_tensor(image, name='image')
if central_fraction <= 0.0 or central_fraction > 1.0:
raise ValueError('central_fraction must be within (0, 1]')
if central_fraction == 1.0:
return image
image = control_flow_ops.with_dependencies(
_Check3DImage(image, require_static=False), image)
img_shape = array_ops.shape(image)
depth = image.get_shape()[2]
img_h = math_ops.to_double(img_shape[0])
img_w = math_ops.to_double(img_shape[1])
bbox_h_start = math_ops.to_int32((img_h - img_h * central_fraction) / 2)
bbox_w_start = math_ops.to_int32((img_w - img_w * central_fraction) / 2)
bbox_h_size = img_shape[0] - bbox_h_start * 2
bbox_w_size = img_shape[1] - bbox_w_start * 2
bbox_begin = array_ops.stack([bbox_h_start, bbox_w_start, 0])
bbox_size = array_ops.stack([bbox_h_size, bbox_w_size, -1])
image = array_ops.slice(image, bbox_begin, bbox_size)
# The first two dimensions are dynamic and unknown.
image.set_shape([None, None, depth])
return image
Example #21
| Source File: metrics_impl.py From auto-alt-text-lambda-api with MIT License | 4 votes |
|
def _sparse_false_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false positive counts.
"""
with ops.name_scope(
None, 'false_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fp = sets.set_size(sets.set_difference(
predictions_idx, labels, aminusb=True))
fp = math_ops.to_double(fp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fp),)):
weights = math_ops.to_double(weights)
fp = math_ops.multiply(fp, weights)
return fp
Example #22
| Source File: metrics_impl.py From lambda-packs with MIT License | 4 votes |
|
def _streaming_sparse_true_positive_at_k(labels,
predictions_idx,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incompatible shape.
"""
with ops.name_scope(
name, _at_k_name('true_positive', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
tp = _sparse_true_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_tp = math_ops.to_double(math_ops.reduce_sum(tp))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_tp, name='update')
Example #23
| Source File: metrics_impl.py From lambda-packs with MIT License | 4 votes |
|
def _sparse_false_negative_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false negative counts.
"""
with ops.name_scope(
None, 'false_negatives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fn = sets.set_size(sets.set_difference(predictions_idx,
labels,
aminusb=False))
fn = math_ops.to_double(fn)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fn),)):
weights = math_ops.to_double(weights)
fn = math_ops.multiply(fn, weights)
return fn
Example #24
| Source File: metrics_impl.py From lambda-packs with MIT License | 4 votes |
|
def _streaming_sparse_false_negative_at_k(labels,
predictions_idx,
k,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incompatible shape.
"""
with ops.name_scope(
name, _at_k_name('false_negative', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
fn = _sparse_false_negative_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_fn = math_ops.to_double(math_ops.reduce_sum(fn))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_fn, name='update')
Example #25
| Source File: metrics_impl.py From lambda-packs with MIT License | 4 votes |
|
def _sparse_false_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false positives for precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false positive counts.
"""
with ops.name_scope(
None, 'false_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fp = sets.set_size(sets.set_difference(
predictions_idx, labels, aminusb=True))
fp = math_ops.to_double(fp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fp),)):
weights = math_ops.to_double(weights)
fp = math_ops.multiply(fp, weights)
return fp
Example #26
| Source File: metrics_impl.py From auto-alt-text-lambda-api with MIT License | 4 votes |
|
def _sparse_true_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None,
name=None):
"""Calculates true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of operation.
Returns:
A [D1, ... DN] `Tensor` of true positive counts.
"""
with ops.name_scope(
name, 'true_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(
labels, predictions_idx, class_id)
tp = sets.set_size(sets.set_intersection(predictions_idx, labels))
tp = math_ops.to_double(tp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, tp),)):
weights = math_ops.to_double(weights)
tp = math_ops.multiply(tp, weights)
return tp
Example #27
| Source File: metrics_impl.py From auto-alt-text-lambda-api with MIT License | 4 votes |
|
def _streaming_sparse_true_positive_at_k(labels,
predictions_idx,
k=None,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
with ops.name_scope(
name, _at_k_name('true_positive', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
tp = _sparse_true_positive_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_tp = math_ops.to_double(math_ops.reduce_sum(tp))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_tp, name='update')
Example #28
| Source File: metrics_impl.py From auto-alt-text-lambda-api with MIT License | 4 votes |
|
def _sparse_false_negative_at_k(labels,
predictions_idx,
class_id=None,
weights=None):
"""Calculates false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
Returns:
A [D1, ... DN] `Tensor` of false negative counts.
"""
with ops.name_scope(
None, 'false_negatives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(labels,
predictions_idx,
class_id)
fn = sets.set_size(sets.set_difference(predictions_idx,
labels,
aminusb=False))
fn = math_ops.to_double(fn)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, fn),)):
weights = math_ops.to_double(weights)
fn = math_ops.multiply(fn, weights)
return fn
Example #29
| Source File: metrics_impl.py From auto-alt-text-lambda-api with MIT License | 4 votes |
|
def _streaming_sparse_false_negative_at_k(labels,
predictions_idx,
k,
class_id=None,
weights=None,
name=None):
"""Calculates weighted per step false negatives for recall@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels`.
If `weights` is `None`, weights default to 1. Use weights of 0 to mask values.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
k: Integer, k for @k metric. This is only used for default op name.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of new variable, and namespace for other dependent ops.
Returns:
A tuple of `Variable` and update `Operation`.
Raises:
ValueError: If `weights` is not `None` and has an incomptable shape.
"""
with ops.name_scope(
name, _at_k_name('false_negative', k, class_id=class_id),
(predictions_idx, labels, weights)) as scope:
fn = _sparse_false_negative_at_k(
predictions_idx=predictions_idx, labels=labels, class_id=class_id,
weights=weights)
batch_total_fn = math_ops.to_double(math_ops.reduce_sum(fn))
var = _local_variable(array_ops.zeros([], dtype=dtypes.float64), name=scope)
return var, state_ops.assign_add(var, batch_total_fn, name='update')
Example #30
| Source File: metrics_impl.py From Serverless-Deep-Learning-with-TensorFlow-and-AWS-Lambda with MIT License | 4 votes |
|
def _sparse_true_positive_at_k(labels,
predictions_idx,
class_id=None,
weights=None,
name=None):
"""Calculates true positives for recall@k and precision@k.
If `class_id` is specified, calculate binary true positives for `class_id`
only.
If `class_id` is not specified, calculate metrics for `k` predicted vs
`n` label classes, where `n` is the 2nd dimension of `labels_sparse`.
Args:
labels: `int64` `Tensor` or `SparseTensor` with shape
[D1, ... DN, num_labels], where N >= 1 and num_labels is the number of
target classes for the associated prediction. Commonly, N=1 and `labels`
has shape [batch_size, num_labels]. [D1, ... DN] must match
`predictions_idx`.
predictions_idx: 1-D or higher `int64` `Tensor` with last dimension `k`,
top `k` predicted classes. For rank `n`, the first `n-1` dimensions must
match `labels`.
class_id: Class for which we want binary metrics.
weights: `Tensor` whose rank is either 0, or n-1, where n is the rank of
`labels`. If the latter, it must be broadcastable to `labels` (i.e., all
dimensions must be either `1`, or the same as the corresponding `labels`
dimension).
name: Name of operation.
Returns:
A [D1, ... DN] `Tensor` of true positive counts.
"""
with ops.name_scope(
name, 'true_positives', (predictions_idx, labels, weights)):
labels, predictions_idx = _maybe_select_class_id(
labels, predictions_idx, class_id)
tp = sets.set_size(sets.set_intersection(predictions_idx, labels))
tp = math_ops.to_double(tp)
if weights is not None:
with ops.control_dependencies((
weights_broadcast_ops.assert_broadcastable(weights, tp),)):
weights = math_ops.to_double(weights)
tp = math_ops.multiply(tp, weights)
return tp
