提交 7ddca5a0 authored 作者: Frederic Bastien's avatar Frederic Bastien

added a first draft of the change log for Theano 0.3

上级 70de3a56
2010-10-22 Theano 0.3
* This is in important release since 0.1. Their have been a 0.2 started internally that was never advertised.
* Their is so much change that we lost trac of many of them. Here is some of them.
* Gpu code with cuda are now generated for many op
* Some interface change since 0.1
* New SharedVariable system to allow reusing memory space between theano function.
* A new memory contract have been done for Theano for people that want to minimize memory copy.
* We Deprecate the old Module system
* No silent downcast of a theano function input by default.
* Raise an error when using the result of logical operation on Theano varible in if (call to __nonzeros__)
* Raise an error when we receive not aligned ndarray.(this is not supported)
* Raise an error when DimShuffle receive bad configuration number.
* Call numpy blas binding for gemv (We already linked with blas gemm)
* If gcc is not enable by default fall back to the python version with a warning.
* Raise an error when tensor.grad is called on a non scalar theano variable
* In the past we where implecitly doing a sum on the tensor to make it a scalar
* added erf and erfc fct.
* Deprecate the current default value of the paramter axis of theano.{max,min,argmax,argmin,max_and_argmax}. We will use the same interface as numpy
* make easy_install theano work
Markdown 格式
0%
您添加了 0 到此讨论。请谨慎行事。
请先完成此评论的编辑!
注册 或者 后发表评论