singa-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF subversion and git services (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SINGA-186) Create Python Tensor class
Date Fri, 17 Jun 2016 14:29:05 GMT

    [ https://issues.apache.org/jira/browse/SINGA-186?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15336210#comment-15336210
] 

ASF subversion and git services commented on SINGA-186:
-------------------------------------------------------

Commit 2a582df121d8b3fa45dad1e9c0435bc01bf79bea in incubator-singa's branch refs/heads/dev
from chonho
[ https://git-wip-us.apache.org/repos/asf?p=incubator-singa.git;h=2a582df ]

SINGA-186 Create Python Tensor class

- Add Swig interfaces: singa.i
  . core_tensor.i for singa::Tensor
  . core_device.i for singa::Device
- Add generate_singa_wrapper.sh to generate wrapper
- Add python script: tensor.py, device.py
- Add an example unittest unittest_python.py in test/python

Notes:
- call copy constructor of singa::Tensor in __init__
- rename transpose -> is_tranpose(), which calls Tensor.transpose()
  rename matrix_transpose -> transpose(), which calls singa::Tensor.T()
- Revised a way to generate Tensor copy in python
  . add copy() for shallow copy, which calls singa::Tensor(Tensor&)
  . add deepcopy() for deep copy, which calls singa::Tensor::Clone()
  Note: revised __init__()
        now clone() and deepcopy() are the same

TODO:
- need think about more efficient way to convert Tensor data to numpy array
- example test is in test/python/example_test_device.py, which will be included in unittest_python.py


> Create Python Tensor class
> --------------------------
>
>                 Key: SINGA-186
>                 URL: https://issues.apache.org/jira/browse/SINGA-186
>             Project: Singa
>          Issue Type: New Feature
>            Reporter: wangwei
>
> This ticket would add a Python Tensor class, which provides functions as the C++ Tensor
class with APIs in Python convention. Most of these functions are implemented by calling the
C++ Tensor functions via Swig.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message