3 ms·

TensorFuse – Common interface for Theano, CGT, and TensorFlow

54 points by dementrock 11y ago

wodenokoto 11y ago
It would have been cool if they posted benchmarks to the different interfaces. But for people wondering, it appears to be a Theano API for interfacing with Tensorflow, and a few other tensor/deep learning libraries. Pretty cool idea.
dancsi 11y ago
Keras (http://keras.io http://keras.io) is a similar library, although only for TensorFlow and Theano...
- shmel 11y ago
  Keras is a very high-level library compare to tensorfuse. Tensorfuse is primarily built to easy migration from theano to something else.
  - Fede_V 11y ago
    Yeah, but Keras has a lower level abstraction layer to switch between Theano and Tensorflow back ends.
  - Smerity 11y ago
    Keras has a low-level compatibility library[1] (`from keras import backend as K`) that people have reported as useful independently. The interface offered by K[2] seems quite similar to that of TensorFuse[3] As the K backend is used for Keras, it also offers proof that K can be used for sizable and complex projects, plus can take advantage of pre-existing testing for Keras. Whilst I love the idea of CGT, it has not yet taken off. I'd be far more interested in seeing a Neon[4] backend considering it has the fastest performance across the board on existing hardware and they're planning to release their own hardware soon. [1]: http://keras.io/backend/ http://keras.io/backend/ [2]: https://github.com/fchollet/keras/blob/master/keras/backend/tensorflow_backend.py https://github.com/fchollet/keras/blob/master/keras/backend/... [3]: https://github.com/dementrock/tensorfuse/tree/master/tensorfuse/backend/tensorflow/tensor https://github.com/dementrock/tensorfuse/tree/master/tensorf... [4]: https://github.com/NervanaSystems/neon https://github.com/NervanaSystems/neon Edit: Incorrectly thought TensorFuse didn't support RNNs, thanks dementrock! Also excited that Lasagne has someone working on being backend independent!
    - dementrock 11y ago
      Actually TensorFuse does support RNN by porting a subset of scan to TensorFlow. I've been using it to port Lasagne to support TensorFlow: https://github.com/dementrock/Lasagne-tf https://github.com/dementrock/Lasagne-tf. The examples/recurrent.py there actually works. Good suggestion on Neon!
mrdrozdov 11y ago
JQuery for scientific frameworks?

54 points by dementrock 11y ago

It would have been cool if they posted benchmarks to the different interfaces. But for people wondering, it appears to be a Theano API for interfacing with Tensorflow, and a few other tensor/deep learning libraries. Pretty cool idea.

dancsi 11y ago

Keras (http://keras.io http://keras.io) is a similar library, although only for TensorFlow and Theano...

shmel 11y ago
Keras is a very high-level library compare to tensorfuse. Tensorfuse is primarily built to easy migration from theano to something else.
- Fede_V 11y ago
  Yeah, but Keras has a lower level abstraction layer to switch between Theano and Tensorflow back ends.
- Smerity 11y ago
  Keras has a low-level compatibility library[1] (`from keras import backend as K`) that people have reported as useful independently. The interface offered by K[2] seems quite similar to that of TensorFuse[3] As the K backend is used for Keras, it also offers proof that K can be used for sizable and complex projects, plus can take advantage of pre-existing testing for Keras. Whilst I love the idea of CGT, it has not yet taken off. I'd be far more interested in seeing a Neon[4] backend considering it has the fastest performance across the board on existing hardware and they're planning to release their own hardware soon. [1]: http://keras.io/backend/ http://keras.io/backend/ [2]: https://github.com/fchollet/keras/blob/master/keras/backend/tensorflow_backend.py https://github.com/fchollet/keras/blob/master/keras/backend/... [3]: https://github.com/dementrock/tensorfuse/tree/master/tensorfuse/backend/tensorflow/tensor https://github.com/dementrock/tensorfuse/tree/master/tensorf... [4]: https://github.com/NervanaSystems/neon https://github.com/NervanaSystems/neon Edit: Incorrectly thought TensorFuse didn't support RNNs, thanks dementrock! Also excited that Lasagne has someone working on being backend independent!
  - dementrock 11y ago
    Actually TensorFuse does support RNN by porting a subset of scan to TensorFlow. I've been using it to port Lasagne to support TensorFlow: https://github.com/dementrock/Lasagne-tf https://github.com/dementrock/Lasagne-tf. The examples/recurrent.py there actually works. Good suggestion on Neon!

mrdrozdov 11y ago

JQuery for scientific frameworks?