From dev-return-4906-archive-asf-public=cust-asf.ponee.io@singa.apache.org Mon Apr 6 03:21:00 2020 Return-Path: X-Original-To: archive-asf-public@cust-asf.ponee.io Delivered-To: archive-asf-public@cust-asf.ponee.io Received: from mail.apache.org (hermes.apache.org [207.244.88.153]) by mx-eu-01.ponee.io (Postfix) with SMTP id 59F40180638 for ; Mon, 6 Apr 2020 05:21:00 +0200 (CEST) Received: (qmail 66533 invoked by uid 500); 6 Apr 2020 03:20:59 -0000 Mailing-List: contact dev-help@singa.apache.org; run by ezmlm Precedence: bulk List-Help: List-Unsubscribe: List-Post: List-Id: Reply-To: dev@singa.apache.org Delivered-To: mailing list dev@singa.apache.org Received: (qmail 66510 invoked by uid 99); 6 Apr 2020 03:20:59 -0000 Received: from ec2-52-202-80-70.compute-1.amazonaws.com (HELO gitbox.apache.org) (52.202.80.70) by apache.org (qpsmtpd/0.29) with ESMTP; Mon, 06 Apr 2020 03:20:58 +0000 From: GitBox To: dev@singa.apache.org Subject: [GitHub] [singa] nudles commented on issue #651: Add new example APIs Message-ID: <158614325889.2301.6560321654505903791.gitbox@gitbox.apache.org> References: In-Reply-To: Date: Mon, 06 Apr 2020 03:20:58 -0000 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit nudles commented on issue #651: Add new example APIs URL: https://github.com/apache/singa/pull/651#issuecomment-609545036 > > > > can we run (cnn|resnet|xceptionnet) + (cifar10|mnist) ? > > I think we can also add the mlp model? > > then we can remove resnet_module.py, mlp_module.py > > There are 4 points, I guess the issue most difficult to solve is the no.1 below > > 1. xceptionnet has some problem, the running stopped after 1 epoch, maybe the model has some issues. And this model is very large I can only run batch size=16 (not 32) what is the error? if you put it on colab, we can debug together. > > 2. The current mlp is 2d input, which is not CNN problem which has 4d input (batch, channel/RGB, height, width). Maybe need to change to 4d when we load the data, pass an argument to specify the dimension (is_2d=True/False). then reshape the data according to the argument inside the `load_data` function. > > 3. The CNN has only one input channel, in order to support cifar10, maybe I need to adjust the number of input channel of CNN as variable > change the CNN model to accept 4d input > 4. mnist is not RGB, so I has only one colour channel, if I used it for resnet I can see if: > (i) change the input channel of resnet to a variable so it support 1 channel > (ii) replicate one channel of mnist dataset to three reshape it into 4d inside the `load_data` function. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: users@infra.apache.org With regards, Apache Git Services