m53297601
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add implicit conversion description for some ops in API.

Browse Source
pull/5314/head
liuxiao93 5 years ago
parent 0c5f7377bd
commit cb67d7d5e6
2 changed files with 162 additions and 1 deletions
Show Stats Download Patch File Download Diff File

55
mindspore/ops/operations/array_ops.py
Unescape View File

@ -2528,6 +2528,11 @@ class ScatterUpdate(_ScatterOp):
Using given values to update tensor value, along with the input indices.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: True.
@ -2569,6 +2574,11 @@ class ScatterNdUpdate(_ScatterNdOp):
Using given values to update tensor value, along with the input indices.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: True.
@ -2610,6 +2620,11 @@ class ScatterMax(_ScatterOp):
Using given values to update tensor value through the max operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: True.
@ -2645,6 +2660,11 @@ class ScatterMin(_ScatterOp):
Using given values to update tensor value through the min operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2674,6 +2694,11 @@ class ScatterAdd(_ScatterOp):
Using given values to update tensor value through the add operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2703,6 +2728,11 @@ class ScatterSub(_ScatterOp):
Using given values to update tensor value through the sub operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2732,6 +2762,11 @@ class ScatterMul(_ScatterOp):
Using given values to update tensor value through the mul operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2761,6 +2796,11 @@ class ScatterDiv(_ScatterOp):
Using given values to update tensor value through the div operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2790,6 +2830,11 @@ class ScatterNdAdd(_ScatterNdOp):
Using given values to update tensor value through the add operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2819,6 +2864,11 @@ class ScatterNdSub(_ScatterNdOp):
Using given values to update tensor value through the sub operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether protect the assignment by a lock. Default: False.
@ -2848,6 +2898,11 @@ class ScatterNonAliasingAdd(_ScatterNdOp):
Using given values to update tensor value through the add operation, along with the input indices.
This operation outputs the `input_x` after the update is done, which makes it convenient to use the updated value.
Inputs of `input_x` and `updates` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **input_x** (Parameter) - The target parameter. The data type should be float16, float32 or int32.
- **indices** (Tensor) - The index to do add operation whose data type should be mindspore.int32.

108
mindspore/ops/operations/nn_ops.py
Unescape View File

@ -1694,6 +1694,12 @@ class ApplyMomentum(PrimitiveWithInfer):
Refer to the paper `On the importance of initialization and momentum in deep
learning <https://dl.acm.org/doi/10.5555/3042817.3043064>`_ for more details.
Inputs of `variable`, `accumulation` and `gradient` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
Data type conversion of Parameter is not supported. RuntimeError exception will be thrown.
Args:
use_locking (bool): Enable a lock to protect the update of variable and accumlation tensors. Default: False.
use_nesterov (bool): Enable Nesterov momentum. Default: False.
@ -3076,6 +3082,11 @@ class FusedSparseAdam(PrimitiveWithInfer):
`beta2_power`, :math:`\alpha` represents `learning_rate`, :math:`w` represents `var`, :math:`\epsilon` represents
`epsilon`.
All of inputs except `indices` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether to enable a lock to protect updating variable tensors.
If true, updates of the var, m, and v tensors will be protected by a lock.
@ -3210,6 +3221,11 @@ class FusedSparseLazyAdam(PrimitiveWithInfer):
`beta2_power`, :math:`\alpha` represents `learning_rate`, :math:`w` represents `var`, :math:`\epsilon` represents
`epsilon`.
All of inputs except `indices` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): Whether to enable a lock to protect updating variable tensors.
If true, updates of the var, m, and v tensors will be protected by a lock.
@ -3325,6 +3341,11 @@ class FusedSparseFtrl(PrimitiveWithInfer):
"""
Merge the duplicate value of the gradient and then update relevant entries according to the FTRL-proximal scheme.
All of inputs except `indices` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
lr (float): The learning rate value, must be positive.
l1 (float): l1 regularization strength, must be greater than or equal to zero.
@ -3423,6 +3444,11 @@ class FusedSparseProximalAdagrad(PrimitiveWithInfer):
.. math::
var = \frac{sign(\text{prox_v})}{1 + lr * l2} * \max(\left| \text{prox_v} \right| - lr * l1, 0)
All of inputs except `indices` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): If true, updates of the var and accum tensors will be protected. Default: False.
@ -3669,6 +3695,12 @@ class ApplyAdaMax(PrimitiveWithInfer):
:math:`beta_1^t` represent `beta1_power`, :math:`var` represents Variable to be updated,
:math:`\epsilon` represents `epsilon`.
Inputs of `var`, `m`, `v` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **var** (Parameter) - Variable to be updated. With float32 or float16 data type.
- **m** (Parameter) - The 1st moment vector in the updating formula. Has the same shape and type as `var`.
@ -3791,6 +3823,12 @@ class ApplyAdadelta(PrimitiveWithInfer):
.. math::
var -= lr * update
Inputs of `var`, `accum`, `accum_update` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **var** (Parameter) - Weights to be updated. With float32 or float16 data type.
- **accum** (Parameter) - Accum to be updated, has the same shape and type as `var`.
@ -3888,6 +3926,12 @@ class ApplyAdagrad(PrimitiveWithInfer):
.. math::
var -= lr * grad * \frac{1}{\sqrt{accum}}
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent..
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
update_slots (bool): If `True`, `accum` will be updated. Default: True.
@ -3963,6 +4007,12 @@ class ApplyAdagradV2(PrimitiveWithInfer):
.. math::
var -= lr * grad * \frac{1}{\sqrt{accum} + \epsilon}
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
epsilon (float): A small value added for numerical stability.
update_slots (bool): If `True`, `accum` will be updated. Default: True.
@ -4040,6 +4090,12 @@ class SparseApplyAdagrad(PrimitiveWithInfer):
.. math::
var -= lr * grad * (1 / sqrt(accum))
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
lr (float): Learning rate.
update_slots (bool): If `True`, `accum` will be updated. Default: True.
@ -4119,6 +4175,12 @@ class SparseApplyAdagradV2(PrimitiveWithInfer):
.. math::
var -= lr * grad * \frac{1}{\sqrt{accum} + \epsilon}
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
lr (float): Learning rate.
epsilon (float): A small value added for numerical stability.
@ -4202,6 +4264,12 @@ class ApplyProximalAdagrad(PrimitiveWithInfer):
.. math::
var = \frac{sign(\text{prox_v})}{1 + lr * l2} * \max(\left| \text{prox_v} \right| - lr * l1, 0)
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): If true, updates of the var and accum tensors will be protected. Default: False.
@ -4298,6 +4366,12 @@ class SparseApplyProximalAdagrad(PrimitiveWithInfer):
.. math::
var = \frac{sign(\text{prox_v})}{1 + lr * l2} * \max(\left| \text{prox_v} \right| - lr * l1, 0)
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
use_locking (bool): If true, updates of the var and accum tensors will be protected. Default: False.
@ -4311,7 +4385,6 @@ class SparseApplyProximalAdagrad(PrimitiveWithInfer):
- **l2** (Union[Number, Tensor]) - l2 regularization strength. should be a float number or
a scalar tensor with float16 or float32 data type..
- **grad** (Tensor) - A tensor of the same type as `var`, for the gradient.
The data type must be float16 or float32.
- **indices** (Tensor) - A vector of indices into the first dimension of `var` and `accum`.
Outputs:
@ -4390,6 +4463,12 @@ class ApplyAddSign(PrimitiveWithInfer):
:math:`t` represents updating step while, :math:`m` represents the 1st moment vector, :math:`m_{t-1}`
is the last momentent of :math:`m_{t}`, :math:`lr` represents scaling factor `lr`, :math:`g` represents `grad`.
Inputs of `var`, `accum` and `grad` comply with the implicit type conversion rules
to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **var** (Parameter) - Variable tensor to be updated. With float32 or float16 data type.
- **m** (Parameter) - Variable tensor to be updated. Has the same dtype as `var`.
@ -4491,6 +4570,13 @@ class ApplyPowerSign(PrimitiveWithInfer):
:math:`t` represents updating step while, :math:`m` represents the 1st moment vector, :math:`m_{t-1}`
is the last momentent of :math:`m_{t}`, :math:`lr` represents scaling factor `lr`, :math:`g` represents `grad`.
All of inputs comply with the implicit type conversion rules to make the data types consistent.
If `lr`, `logbase`, `sign_decay` or `beta` is a number, the number is automatically converted to Tensor,
and the data type is consistent with the Tensor data type involved in the operation.
If inputs are tensors and have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **var** (Parameter) - Variable tensor to be updated. With float32 or float16 data type.
If data type of `var` is float16, all inputs must have the same data type as `var`.
@ -4587,6 +4673,11 @@ class ApplyGradientDescent(PrimitiveWithInfer):
.. math::
var = var - \alpha * \delta
Inputs of `var` and `delta` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **var** (Parameter) - Variable tensor to be updated. With float32 or float16 data type.
- **alpha** (Union[Number, Tensor]) - Scaling factor, should be a scalar. With float32 or float16 data type.
@ -4649,6 +4740,11 @@ class ApplyProximalGradientDescent(PrimitiveWithInfer):
.. math::
var = \frac{sign(\text{prox_v})}{1 + \alpha * l2} * \max(\left| \text{prox_v} \right| - alpha * l1, 0)
Inputs of `var` and `delta` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Inputs:
- **var** (Parameter) - Variable tensor to be updated. With float32 or float16 data type.
- **alpha** (Union[Number, Tensor]) - Saling factor, should be a scalar. With float32 or float16 data type.
@ -4886,6 +4982,11 @@ class SparseApplyFtrl(PrimitiveWithInfer):
"""
Update relevant entries according to the FTRL-proximal scheme.
All of inputs except `indices` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
lr (float): The learning rate value, must be positive.
l1 (float): l1 regularization strength, must be greater than or equal to zero.
@ -4973,6 +5074,11 @@ class SparseApplyFtrlV2(PrimitiveWithInfer):
"""
Update relevant entries according to the FTRL-proximal scheme.
All of inputs except `indices` comply with the implicit type conversion rules to make the data types consistent.
If they have different data types, lower priority data type will be converted to
relatively highest priority data type.
RuntimeError exception will be thrown when the data type conversion of Parameter is required.
Args:
lr (float): The learning rate value, must be positive.
l1 (float): l1 regularization strength, must be greater than or equal to zero.

Write Preview
Loading…
Cancel
Save