chainer/Lobby

Sameroom

@sameroom-bot

[yixuan, chainer] @crcrpar does there have any guide to build chaine template with cuda 10?

Sameroom

@sameroom-bot

[Masaki Kozuki, chainer] The installation guides of Chainer & CuPy are enough I think.

For cuda10.0 environment, I think the below is enough according to the above guides.

$ pip install -U setuptools pip
$ pip install cupy-cuda100
$ pip install chainer

Sameroom

@sameroom-bot

[yixuan, chainer] thanks for your information. i will try later @crcrpar

Sameroom

@sameroom-bot

[Akshay Subramanian, chainer] Untitled

[Akshay Subramanian, chainer] chainerx.is_available() is also returning False.

Sameroom

@sameroom-bot

[Kshiteej K, chainer] I faced this issue few times when, there was a mismatch in function declaration and definition. Maybe check the code once to see that is correct.

Sameroom

@sameroom-bot

[Wang Yu, chainer] Hello, Beg your help, please. I just want to calculate the cross_entropy between two variables without using softmax, so is there any function available?

[Masaki Kozuki, chainer] Is [F.sigmoid_cross_entropy](https://docs.chainer.org/en/stable/reference/generated/chainer.functions.sigmoid_cross_entropy.html#chainer.functions.sigmoid_cross_entropy) not enough?

Sameroom

@sameroom-bot

[Wang Yu, chainer] Thank you for your answer, but I just want the cross_entropy without any activation function.

Sameroom

@sameroom-bot

[Masaki Kozuki, chainer] OK, then I would implement that function using chainer.functions. :)

Sameroom

@sameroom-bot

[Wang Yu, chainer] Thanks a lot:D

Sameroom

@sameroom-bot

[Anshuk Uppal, chainer] Sorry for troubling you guys with this but I couldn't find anything definite upon googling. Q: Does Chainer support backprop on vector losses? I have been told so before but simple codes do not work, am I missing something here?

Sameroom

@sameroom-bot

[Anshuk Uppal, chainer] As much as I could trace the problem the grad_var being passed in line 1438(variable.py) : _backprop_to_all([(node, grad_var)], retain_grad, loss_scale) is None and this propagates through to the gradient calculation

Sameroom

@sameroom-bot

[Katsuhiko Nishimra, chainer] Why not F.sum(vector).backward()?

Sameroom

@sameroom-bot

[Anshuk Uppal, chainer] the application requires non-accumalated gradients at the links

Sameroom

@sameroom-bot

[Anshuk Uppal, chainer] not higher level links, simpler ones like functions with parameters

Sameroom

@sameroom-bot

[Tommi Kerola, chainer] https://github.com/chainer/chainer/blob/master/chainer/variable.py#L1393

This method uses :data:grad as the initial error array. User can manually set a gradient array before calling this method.
So if you set vector.grad = xp.ones(shape), I think you can backprop vector losses.

Sameroom

@sameroom-bot

[Anshuk Uppal, chainer] Thank you for the reply, I was initially

Sameroom

@sameroom-bot

[Fernando da Motta Hildebrand, chainer] Hi. I never even heard of Chainer until today. I hope this is the right channel( also, sorry for my english). It made me wonder, why is Pytorch so popular? I ask because Im in the beggining of my masters course(bioinfo) and Im experimenting with Keras and Pytorch for a CNN to classify proteins(Dont like pytorch very much tought...). Im a little afraid of frameworks that dont have much community suport. So, could you guys clear those questions for me?

Sameroom

@sameroom-bot

[Seiya Tokui, chainer] It is a difficult question repeatedly raised by us, too! I really want to know the essential factor of their success.

The obvious factors that I came up with are: 1) their location (closer to the top-notch researchers/labs, other important partners like NVIDIA, etc., while Chainer is developed in Japan), 2) strong marketing (I heard there was a closing beta period involving researchers outside Facebook. Their website is also fancy from the beginning, maybe attracting young researchers who are the core of DL community), 3) being backed by a big company. I do not think these points explain everything, though.

Sameroom

@sameroom-bot

[Cloud Han, chainer] How capable is ChainerX now, are there any demo project?

Sameroom

@sameroom-bot

[Cloud Han, chainer] Now I need a not so powerful tensor library. Seems ChainerX is the one to go. I have investigated PyTorch's ATen (A TENsor library), although the name, it cannot be built as a standalone LIBRARY, so I give up. It seems ChainerX is developing for deploying in mind, is there any easy mechanism disable the back prop code during compiling (to generate a smaller lib in size). Also, after looking into the chainerx_cc directory. I don't find too much cuda code there. But with nm --demangle libchainerx.so I see a bunch of them, which do not seem to be included in CMakeLists.txt. What have I missed? Now I am quiet curious with the compiling process..

[Cloud Han, chainer] As a side note, by disabling the gencode for non using cuda arch, and striping the lib, I obtain a lib in 70MB or so. Great works!

Sameroom

@sameroom-bot

[nishino, chainer] Thank you for taking interest!
This is the tracking issue for ChainerX op coverage. There is a link to a spreadsheet that list up supported ops: chainer/chainer#6423
There is an example code here:
https://github.com/chainer/chainer/tree/master/chainerx_cc/examples/mnist
It's a deep learning example, though.
Currently there's no way to disable backprop code in compilation, but technically it should be doable. It would be great if you could submit a feature request as an issue.
You can find all of the CUDA code under chainerx_cc/chainerx/cuda.

Sameroom

@sameroom-bot

[Saanidhya vats, chainer] Hello everyone, I am saanidhya vats currently pursuing btech from computer science branch. I am new to open source and want to contribute. How shall I start?

Sameroom

@sameroom-bot

[dselivanov, chainer] Hey here. Does switching to pytorch means freezing ChainerX develpment as well?

Sameroom

@sameroom-bot

[Kenichi Maehashi, chainer] Yes, ChainerX is a part of Chainer and further development will be bug-fix and maintenance only.

Sameroom

@sameroom-bot

[Swapnil Kannojia, chainer] Hey everyone ! I am Swapnil Kannojia , an enthusiast in machine learning and deep learning. I want to have more experience in open source development and machine learning projects. Are there any issues I can work on and contribute ?

Sameroom

@sameroom-bot

[Brandon Dube, chainer] What does this mean for ChainerX?

Sameroom

@sameroom-bot

[Leo Fang, chainer] Just curious, have you considered changing Chainer’s project management model, allowing external contributors to join the core dev team so that it can still be kept alive (with new features added in every release, etc)? Maybe some people (not me (joy/allthethings emoji)) would love to direct the future development of Chainer?

Sameroom

@sameroom-bot

[UmashankarTriforce, chainer] Happy New Year everyone!

Sameroom

@sameroom-bot

[Bibek Behera, chainer] hi is chainer built on top of transformer

Sameroom

@sameroom-bot

[Bibek Behera, chainer] Sorry i meant to say tensorflow

Sameroom

@sameroom-bot

[Emilio, chainer] No, chainer is a complete framework built from 0.
We have ceased to add new functionalities to it although we are doing maintenance.

Sameroom

@sameroom-bot

[UmashankarTriforce, chainer] As the registrations for GSoC organizations has been opened, has CuPy decided to participate in GSoC this year?

Sameroom

@sameroom-bot

[UmashankarTriforce, chainer] As the registrations for GSoC organizations has been opened, has CuPy decided to participate in GSoC this year?

Sameroom

@sameroom-bot

[Emilio, chainer] Hello, sorry we still have not decided if we will participate this year.
If we do so, we will make an announce with the steps to submit your application.
Thanks!

Sameroom

@sameroom-bot

[Leo Fang, chainer] I suppose you don’t need external mentors? :rolling_on_the_floor_laughing:

Sameroom

@sameroom-bot

[UmashankarTriforce, chainer] Any updates on cupy's participation in gsoc this year? If I'm not wrong, organisation registration closes in less than a day.

Sameroom

@sameroom-bot

[Emilio, chainer] We applied to GSoC

[Emilio, chainer] https://github.com/cupy/cupy/wiki/GSoC-2020-Project-Ideas

[Kenichi Maehashi, chainer] Hi, this year we applied to GSoC for CuPy and Optuna project.
https://github.com/numfocus/gsoc#organizations-confirmed-under-numfocus-umbrella
https://github.com/cupy/cupy/wiki/GSoC-2020-Project-Ideas

Sameroom

@sameroom-bot

[UmashankarTriforce, chainer] This is great news! Can we start discussing with mentors about the projects?

Sameroom

@sameroom-bot

[Kenichi Maehashi, chainer] I made a channel for discussion related to GSoC 2020 #gsoc20 . Please join if you have interest!

Sameroom

@sameroom-bot

[Albert Kahira, chainer] Is there a way to make Dropout layer deterministic without using mask/

Rushabh Vasani

@rushabh-v

I am not able to join #gsoc20 channel. How to join that?

Sameroom

@sameroom-bot

[Masaki Kozuki, chainer] You might need to join slack. Here’s the invitation link. https://chainer.slack.com/join/shared_invite/enQtMjk4ODE3NjA4MzM3LTMzNTk5NjFkOTBmYmE1YjAwMDAzMDBhN2I3ODEzZjA3ODZjZDg1MDlhYjI0MDFiYThmNGIwM2FiNTljNDJmMmI

Sameroom

@sameroom-bot

[Ali Torabi, chainer] Hi everybody

[Ali Torabi, chainer] I’ve got a question. When I try to run my code on colab, there is such a frustrating error: AttributeError: module ‘chainer.functions’ has no attribute ‘Concat’. Everything upgraded on latest version but the error remains. Thanks for helping out.

Sameroom

@sameroom-bot

[Albert Kahira, chainer] Is there anyway to use all cores in chainer for training with CPU. At the moment, it seems to use only one core

Where communities thrive

People

Repo info

Activity