diff --git a/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_api_mapping.md b/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_api_mapping.md index ca1fce5b6e7d43bc28cfd8baa295bab8d5af67d4..13000bf4b8e66937a65cc26c4df088ced2a34e33 100644 --- a/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_api_mapping.md +++ b/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_api_mapping.md @@ -79,3 +79,16 @@ Mapping between TensorFlow APIs and MindSpore APIs, which is provided by the com | [tf.zeros_like](http://www.tensorflow.org/versions/r1.15/api_docs/python/tf/zeros_like) |[mindspore.ops.ZerosLike](https://mindspore.cn/docs/api/en/master/api_python/ops/mindspore.ops.ZerosLike.html) | same | | [tf.distribute.Strategy](http://www.tensorflow.org/versions/r1.15/api_docs/python/tf/distribute/Strategy) |[context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL)](https://mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.context.html#mindspore.context.set_auto_parallel_context) | [diff](https://www.mindspore.cn/docs/migration_guide/en/master/api_mapping/tensorflow_diff/DistributedTrain.html)| + + +# TensorFlow Probability and MindSpore + + + +Mapping between TensorFlow Probability APIs and MindSpore APIs, which is provided by the community. + +| TensorFlow Probability 0.14.1 APIs | MindSpore APIs + | Description | +|-----------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------------------|------| + +| [tfp.bijectors.Softplus](http://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/Softplus) |[mindspore.nn.probability.bijector.Softplus](https://www.mindspore.cn/docs/api/en/master/api_python/nn_probability/mindspore.nn.probability.bijector.Softplus.html)|[diff](https://www.mindspore.cn/docs/migration_guide/en/master/api_mapping/tensorflow_probability_diff/BijectorSoftplus.html)| diff --git a/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_probability_diff/BijectorSoftplus.md b/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_probability_diff/BijectorSoftplus.md new file mode 100644 index 0000000000000000000000000000000000000000..703da8f3809a5b32fb16bd2614ecb3554d523a7e --- /dev/null +++ b/docs/mindspore/migration_guide/source_en/api_mapping/tensorflow_probability_diff/BijectorSoftplus.md @@ -0,0 +1,88 @@ +# Function Differences with tfp.bijectors.Softplus + + + +## tfp.bijectors.Softplus + +```python +class tfp.bijectors.Softplus( + hinge_softness=None, + low=None, + validate_args=False, + name='softplus' +) +``` + +For more information, see[tfp.bijectors.Softplus](http://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/Softplus). + +## mindspore.nn.probability.bijector.Softplus + +```python +class mindspore.nn.probability.bijector.Softplus( + sharpness=1.0, + name="Softplus" +) +``` + +For more information, see[mindspore.nn.probability.bijector.Softplus](https://www.mindspore.cn/docs/api/en/master/api_python/nn_probability/mindspore.nn.probability.bijector.Softplus.html). + +## Differences + +TensorFlow: The formula: Y = g(X) = Log[1 + exp(X)]. + +MindSpore：The formula: $Y = g(X) = log(1 + e ^ {kX}) / k $. Therefore, when sharpness = 1./hinge_softness, the calculation results of mindspore and tensorflow are equal. + +## Code Example + +```python +# The following implements bijector.Softplus with MindSpore. +import tensorflow as tf +import tensorflow_probability.python as tfp +import mindspore +import mindspore.nn as nn +from mindspore import Tensor +import mindspore.nn.probability.bijector as msb + +# To initialize a Softplus bijector of sharpness 2.0. +softplus = msb.Softplus(2.0) +value = Tensor([2], dtype=mindspore.float32) +ans1 = softplus.forward(value) +print(ans1) +#Out: +#[2.009075] +ans2 = softplus.inverse(value) +print(ans2) +#Out: +#[1.9907573] +ans3 = softplus.forward_log_jacobian(value) +print(ans3) +#Out: +#[-0.01814996] +ans4 = softplus.inverse_log_jacobian(value) +print(ans4) +#Out: +#[0.01848531] + + +# The following implements bijectors.Softplus with TensorFlow_Probability. +value_tf = tf.constant([2], dtype=tf.float32) +# sharpness = 2.0, sharpness = 1./hinge_softness, so hinge_softness = 0.5 +output = tfp.bijectors.Softplus(0.5) +out1 = output.forward(value_tf) +out2 = output.inverse(value_tf) +out3 = output.forward_log_det_jacobian(value_tf, event_ndims=0) +out4 = output.inverse_log_det_jacobian(value_tf, event_ndims=0) +ss = tf.Session() +ss.run(out1) +# out1 +# array([2.009075], dtype=float32) +ss.run(out2) +# out2 +# array([1.9907573], dtype=float32) +ss.run(out3) +# out3 +# array([-0.01814996], dtype=float32) +ss.run(out4) +# out4 +# array([0.01848542], dtype=float32) +``` diff --git a/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_api_mapping.md b/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_api_mapping.md index d22da72bc139fab9d42c1a328353841c1b4e5cd1..0e7870c5ea2acdba53cef991bb8909a31818f513 100644 --- a/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_api_mapping.md +++ b/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_api_mapping.md @@ -80,3 +80,13 @@ | [tf.distribute.Strategy](http://www.tensorflow.org/versions/r1.15/api_docs/python/tf/distribute/Strategy) |[context.set_auto_parallel_context(parallel_mode=ParallelMode.DATA_PARALLEL)](https://mindspore.cn/docs/api/zh-CN/master/api_python/mindspore.context.html#mindspore.context.set_auto_parallel_context) | [差异对比](https://www.mindspore.cn/docs/migration_guide/zh-CN/master/api_mapping/tensorflow_diff/DistributedTrain.html)| + +# TensorFlow Probability与MindSpore + + + +由社区提供的TensorFlow Probability APIs和MindSpore APIs之间的映射。 + +| TensorFlow Probability 0.14.1 APIs | MindSpore APIs | 说明 | +|-----------------------------------------------------------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------| +| [tfp.bijectors.Softplus](http://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/Softplus) |[mindspore.nn.probability.bijector.Softplus](https://www.mindspore.cn/docs/api/zh-CN/master/api_python/nn_probability/mindspore.nn.probability.bijector.Softplus.html)|[差异对比](https://www.mindspore.cn/docs/migration_guide/zh-CN/master/api_mapping/tensorflow_probability_diff/BijectorSoftplus.html)| diff --git a/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_probability_diff/BijectorSoftplus.md b/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_probability_diff/BijectorSoftplus.md new file mode 100644 index 0000000000000000000000000000000000000000..f350f29fb1ea7162a4f5f12ef800687b9f56eea0 --- /dev/null +++ b/docs/mindspore/migration_guide/source_zh_cn/api_mapping/tensorflow_probability_diff/BijectorSoftplus.md @@ -0,0 +1,88 @@ +# 比较与tfp.bijectors.Softplus的功能差异 + + + +## tfp.bijectors.Softplus + +```python +class tfp.bijectors.Softplus( + hinge_softness=None, + low=None, + validate_args=False, + name='softplus' +) +``` + +更多内容详见[tfp.bijectors.Softplus](http://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/Softplus)。 + +## mindspore.nn.probability.bijector.Softplus + +```python +class mindspore.nn.probability.bijector.Softplus( + sharpness=1.0, + name="Softplus" +) +``` + +更多内容详见[mindspore.nn.probability.bijector.Softplus](https://www.mindspore.cn/docs/api/zh-CN/master/api_python/nn_probability/mindspore.nn.probability.bijector.Softplus.html)。 + +## 使用方式 + +TensorFlow：公式：Y = g(X) = Log[1 + exp(X)] + +MindSpore：公式：$Y = g(X) = log(1 + e ^ {kX}) / k $， 所以当sharpness = 1./hinge_softness的时候，MindSpore与TensorFlow的计算结果是一致的。 + +## 代码示例 + +```python +# The following implements bijector.Softplus with MindSpore. +import tensorflow as tf +import tensorflow_probability.python as tfp +import mindspore +import mindspore.nn as nn +from mindspore import Tensor +import mindspore.nn.probability.bijector as msb + +# To initialize a Softplus bijector of sharpness 2.0. +softplus = msb.Softplus(2.0) +value = Tensor([2], dtype=mindspore.float32) +ans1 = softplus.forward(value) +print(ans1) +#Out: +#[2.009075] +ans2 = softplus.inverse(value) +print(ans2) +#Out: +#[1.9907573] +ans3 = softplus.forward_log_jacobian(value) +print(ans3) +#Out: +#[-0.01814996] +ans4 = softplus.inverse_log_jacobian(value) +print(ans4) +#Out: +#[0.01848531] + + +# The following implements bijectors.Softplus with TensorFlow_Probability. +value_tf = tf.constant([2], dtype=tf.float32) +# sharpness = 2.0, sharpness = 1./hinge_softness, so hinge_softness = 0.5 +output = tfp.bijectors.Softplus(0.5) +out1 = output.forward(value_tf) +out2 = output.inverse(value_tf) +out3 = output.forward_log_det_jacobian(value_tf, event_ndims=0) +out4 = output.inverse_log_det_jacobian(value_tf, event_ndims=0) +ss = tf.Session() +ss.run(out1) +# out1 +# array([2.009075], dtype=float32) +ss.run(out2) +# out2 +# array([1.9907573], dtype=float32) +ss.run(out3) +# out3 +# array([-0.01814996], dtype=float32) +ss.run(out4) +# out4 +# array([0.01848542], dtype=float32) +```