Write a tutorial for deeplearn.js #15

chihuahua · 2017-08-07T09:09:31Z

This tutorial takes the reader through making a model to predict the complements of colors within RGB space. We make a simple feed forward network with 3 fully-connected layers.

This change is

This tutorial takes the reader through making a model to predict the complements of colors within RGB space. We make a simple feed forward network with 3 fully-connected layers.

dsmilkov · 2017-08-08T15:37:20Z

docs/tutorials/complementary_color_prediction.md

@@ -0,0 +1,194 @@
+# Predicting Complementary Colors with deeplearn.js


Keep this file in demos/complementary-color-prediction/ - doesn't have to be in docs/tutorials. The website will pick it up and show it on the right side in https://pair-code.github.io/deeplearnjs/docs/tutorials/index.html automatically

dsmilkov · 2017-08-08T15:40:52Z

demos/complementary-color-prediction/complementary-color-prediction.html

@@ -0,0 +1,136 @@
+<!-- Copyright 2017 Google Inc. All Rights Reserved.


Can you run the demo using watch-demo and share the url with us? Thanks!

dsmilkov · 2017-08-08T15:43:09Z

demos/complementary-color-prediction/complementary-color-prediction.ts

@@ -0,0 +1,355 @@
+/* Copyright 2017 Google Inc. All Rights Reserved.


Reminder to run npm run lint to check for lint errors. Thanks for your patience before we automate this with travis CI.

I get this output

agent007@agent007:~/deeplearnjs$ npm run lint

deeplearn@0.1.0 lint /usr/home/agent007/deeplearnjs
tslint -p . --type-check -t verbose

ERROR: (no-any) /usr/home/agent007/deeplearnjs/demos/homepage/index.ts[30, 63]: Type declaration of 'any' loses type-safety. Consider replacing it with a more precise type, the empty type ('{}'), or suppress this occurrence.
ERROR: (triple-equals) /usr/home/agent007/deeplearnjs/demos/homepage/index.ts[21, 28]: != should be !==

npm ERR! code ELIFECYCLE
npm ERR! errno 2
npm ERR! deeplearn@0.1.0 lint: tslint -p . --type-check -t verbose
npm ERR! Exit status 2
npm ERR!
npm ERR! Failed at the deeplearn@0.1.0 lint script.
npm ERR! This is probably not a problem with npm. There is likely additional logging output above.

npm ERR! A complete log of this run can be found in:
npm ERR! /usr/home/agent007/.npm/_logs/2017-08-09T08_43_14_372Z-debug.log

dsmilkov · 2017-08-11T19:57:42Z

Reviewed 2 of 6 files at r2, 1 of 2 files at r3.
Review status: 3 of 6 files reviewed at latest revision, 7 unresolved discussions.

npm-debug.log, line 1 at r4 (raw file):

0 info it worked if it ends with ok

revert this file

demos/.DS_Store, line 0 at r4 (raw file):
revert this file also

demos/complementary-color-prediction/complementary-color-prediction.html, line 11 at r4 (raw file):

See the License for the specific language governing permissions and
limitations under the License.
==============================================================================-->

reduce last line of license to 80 width

demos/complementary-color-prediction/complementary-color-prediction.ts, line 1 at r1 (raw file):

Previously, chihuahua (Chi Zeng) wrote…

I get this output

agent007@agent007:~/deeplearnjs$ npm run lint

deeplearn@0.1.0 lint /usr/home/agent007/deeplearnjs
tslint -p . --type-check -t verbose

ERROR: (no-any) /usr/home/agent007/deeplearnjs/demos/homepage/index.ts[30, 63]: Type declaration of 'any' loses type-safety. Consider replacing it with a more precise type, the empty type ('{}'), or suppress this occurrence.
ERROR: (triple-equals) /usr/home/agent007/deeplearnjs/demos/homepage/index.ts[21, 28]: != should be !==

npm ERR! code ELIFECYCLE
npm ERR! errno 2
npm ERR! deeplearn@0.1.0 lint: tslint -p . --type-check -t verbose
npm ERR! Exit status 2
npm ERR!
npm ERR! Failed at the deeplearn@0.1.0 lint script.
npm ERR! This is probably not a problem with npm. There is likely additional logging output above.

npm ERR! A complete log of this run can be found in:
npm ERR! /usr/home/agent007/.npm/_logs/2017-08-09T08_43_14_372Z-debug.log

Did you run npm run prep? Also can you check that you have the right version of the tsc compiler (2.3.4) by running node_modules/.bin/tsc --version?

demos/complementary-color-prediction/complementary_color_prediction.md, line 3 at r4 (raw file):

may not be perfectly optimized

maybe elaborate a bit on this. Are you thinking about hyper-params or not optimized for speed (not using the library in the best way)?

If hyper-params, you can say smth like
'We didn't spend time tuning the hyper-parameters so there is still lots of room - we'd love a pull request.'

demos/complementary-color-prediction/complementary_color_prediction.md, line 271 at r4 (raw file):

Based on a few sample colors, it seems like our model of `64 + 32 + 16 = 112` middle-layer nodes fairs alright.

TODO(chihuahua): Update this image to complementary-color-prediction.png once it is checked in.

Remove TODO and update the link now? Since it will get checked in this PR.

demos/complementary-color-prediction/complementary_color_prediction.md, line 274 at r4 (raw file):

offer

s/offer/offers/ (might be wrong, feel free to ignore)

Comments from Reviewable

Also modified .gitignore to avoid re-adding them.

chihuahua · 2017-08-12T01:42:08Z

Review status: 2 of 5 files reviewed at latest revision, 7 unresolved discussions, some commit checks failed.

demos/complementary-color-prediction/complementary-color-prediction.html, line 11 at r4 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

reduce last line of license to 80 width

Done. Moved closing comment tag to the next line.

demos/complementary-color-prediction/complementary-color-prediction.ts, line 1 at r1 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

Did you run npm run prep? Also can you check that you have the right version of the tsc compiler (2.3.4) by running node_modules/.bin/tsc --version?

Indeed, version 2.3.4 ...

agent007@agent007:~/Desktop/deeplearnjs$ node_modules/.bin/tsc --version
Version 2.3.4

I did run npm run prep. I tried again and think I get the same output ... odd. Here is the output of the log file. I might have some time later to dig deeper.

0 info it worked if it ends with ok
1 verbose cli [ '/usr/local/home/agent007/.nvm/versions/node/v8.3.0/bin/node',
1 verbose cli   '/usr/local/home/agent007/.nvm/versions/node/v8.3.0/bin/npm',
1 verbose cli   'run',
1 verbose cli   'lint' ]
2 info using npm@5.3.0
3 info using node@v8.3.0
4 verbose run-script [ 'prelint', 'lint', 'postlint' ]
5 info lifecycle deeplearn@0.1.0~prelint: deeplearn@0.1.0
6 info lifecycle deeplearn@0.1.0~lint: deeplearn@0.1.0
7 verbose lifecycle deeplearn@0.1.0~lint: unsafe-perm in lifecycle true
8 verbose lifecycle deeplearn@0.1.0~lint: PATH: /usr/local/home/agent007/.nvm/versions/node/v8.3.0/lib/node_modules/npm/bin/node-gyp-bin:/usr/local/home/agent007/Desktop/deeplearnjs/node_modules/.bin:/usr/local/home/agent007/.nvm/versions/node/v8.3.0/bin:/usr/local/home/agent007/anaconda2/bin:/usr/lib/google-golang/bin:/usr/local/buildtools/java/jdk/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/src/head/depot/google3/third_party/gpus/cuda_8_0/bin
9 verbose lifecycle deeplearn@0.1.0~lint: CWD: /usr/local/home/agent007/Desktop/deeplearnjs
10 silly lifecycle deeplearn@0.1.0~lint: Args: [ '-c', 'tslint -p . --type-check -t verbose' ]
11 silly lifecycle deeplearn@0.1.0~lint: Returned: code: 2  signal: null
12 info lifecycle deeplearn@0.1.0~lint: Failed to exec lint script
13 verbose stack Error: deeplearn@0.1.0 lint: `tslint -p . --type-check -t verbose`
13 verbose stack Exit status 2
13 verbose stack     at EventEmitter.<anonymous> (/usr/local/home/agent007/.nvm/versions/node/v8.3.0/lib/node_modules/npm/lib/utils/lifecycle.js:289:16)
13 verbose stack     at emitTwo (events.js:125:13)
13 verbose stack     at EventEmitter.emit (events.js:213:7)
13 verbose stack     at ChildProcess.<anonymous> (/usr/local/home/agent007/.nvm/versions/node/v8.3.0/lib/node_modules/npm/lib/utils/spawn.js:40:14)
13 verbose stack     at emitTwo (events.js:125:13)
13 verbose stack     at ChildProcess.emit (events.js:213:7)
13 verbose stack     at maybeClose (internal/child_process.js:927:16)
13 verbose stack     at Process.ChildProcess._handle.onexit (internal/child_process.js:211:5)
14 verbose pkgid deeplearn@0.1.0
15 verbose cwd /usr/local/home/agent007/Desktop/deeplearnjs
16 verbose Linux 4.4.0-83-generic
17 verbose argv "/usr/local/home/agent007/.nvm/versions/node/v8.3.0/bin/node" "/usr/local/home/agent007/.nvm/versions/node/v8.3.0/bin/npm" "run" "lint"
18 verbose node v8.3.0
19 verbose npm  v5.3.0
20 error code ELIFECYCLE
21 error errno 2
22 error deeplearn@0.1.0 lint: `tslint -p . --type-check -t verbose`
22 error Exit status 2
23 error Failed at the deeplearn@0.1.0 lint script.
23 error This is probably not a problem with npm. There is likely additional logging output above.
24 verbose exit [ 2, true ]

demos/complementary-color-prediction/complementary_color_prediction.md, line 3 at r4 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

may not be perfectly optimized

maybe elaborate a bit on this. Are you thinking about hyper-params or not optimized for speed (not using the library in the best way)?

If hyper-params, you can say smth like
'We didn't spend time tuning the hyper-parameters so there is still lots of room - we'd love a pull request.'

Done. Right - I meant that this tutorial did not optimize for hyperparameters (number of layers, sizes of layers). If you have any feedback on using the library better - let me know!

demos/complementary-color-prediction/complementary_color_prediction.md, line 271 at r4 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

Remove TODO and update the link now? Since it will get checked in this PR.

I removed the TODO and verified that the markdown should look correct:

The markdown preview currently does not show the image because the URL is relative to the PAIR-code organization, and the image has not yet been checked into the PAIR-code/deeplearnjs repo yet.

demos/complementary-color-prediction/complementary_color_prediction.md, line 274 at r4 (raw file):

s within it offer a
Done.

Also, to avoid bloating this tutorial, I sort of glossed over the logic for managing the UI. That also makes the tutorial focus on deeplearn.js. Let me know if you have thoughts on that.

npm-debug.log, line 1 at r4 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

revert this file

Done.

demos/.DS_Store, line at r4 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

revert this file also

Done.

Comments from Reviewable

dsmilkov · 2017-08-12T02:12:21Z

Reviewed 3 of 6 files at r2, 1 of 1 files at r4, 5 of 5 files at r5.
Review status: all files reviewed at latest revision, 2 unresolved discussions, some commit checks failed.

demos/complementary-color-prediction/complementary_color_prediction.md, line 271 at r4 (raw file):

Previously, chihuahua (Chi Zeng) wrote…

I removed the TODO and verified that the markdown should look correct:

The markdown preview currently does not show the image because the URL is relative to the PAIR-code organization, and the image has not yet been checked into the PAIR-code/deeplearnjs repo yet.

I think you can do relative paths in markdown links, .e.g
Predicting Complementary Colors
also when linking the other articles from this article.

demos/complementary-color-prediction/complementary_color_prediction.md, line 274 at r4 (raw file):

Previously, chihuahua (Chi Zeng) wrote…

s within it offer a
Done.

Also, to avoid bloating this tutorial, I sort of glossed over the logic for managing the UI. That also makes the tutorial focus on deeplearn.js. Let me know if you have thoughts on that.

I think it's good. Short and sweet. We can get feedback from our users and improve later.

Comments from Reviewable

chihuahua · 2017-08-13T00:03:59Z

Review status: 3 of 5 files reviewed at latest revision, all discussions resolved, some commit checks failed.

demos/complementary-color-prediction/complementary_color_prediction.md, line 271 at r4 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

I think you can do relative paths in markdown links, .e.g
Predicting Complementary Colors
also when linking the other articles from this article.

Indeed, relative links work! Done.

Comments from Reviewable

chihuahua · 2017-08-13T00:18:03Z

I modified relative links. Feel free to merge at any time.

One problem I realized is that, sometimes, certain weight values are initialized to 0, and then those color channels just stay at 0. Like this:

I so far lack a solution for this. Refreshing the page usually allows for normal training, but it's unfortunate that initialization of weights matters.

chihuahua · 2017-08-13T00:33:07Z

Interestingly, despite having a channel stuck at 0, the predictions are often not that bad:

Anyways, I added a new section in the markdown that documents this caveat.

dsmilkov · 2017-08-13T00:55:49Z

Ha, that's very interesting about the weights going to 0.

Wohoo, thanks for writing this awesome tutorial!

* Write a tutorial for deeplearn.js This tutorial takes the reader through making a model to predict the complements of colors within RGB space. We make a simple feed forward network with 3 fully-connected layers. * . * Finish md * Update TOC * Update complementary_color_prediction.md * Remove npm-debug.log and .DS_Store Also modified .gitignore to avoid re-adding them. * Address comments * Change image URL * Move comment closing tag to next line * Make png link relative * Remove unused imports * Make links relative * Describe problem of channel stuck at 0 * . * .

[Mulmat] register flags for matmul

chihuahua added 2 commits August 7, 2017 01:51

Write a tutorial for deeplearn.js

03303d7

This tutorial takes the reader through making a model to predict the complements of colors within RGB space. We make a simple feed forward network with 3 fully-connected layers.

.

46acc79

dsmilkov reviewed Aug 8, 2017

View reviewed changes

chihuahua and others added 3 commits August 9, 2017 01:22

Finish md

6312b6b

Update TOC

0e717f6

Update complementary_color_prediction.md

fffa88b

chihuahua added 4 commits August 11, 2017 17:44

Remove npm-debug.log and .DS_Store

2cb5887

Also modified .gitignore to avoid re-adding them.

Address comments

b1bbcc5

Change image URL

b10c8e6

Move comment closing tag to next line

7d6bf35

chihuahua added 3 commits August 12, 2017 16:56

Make png link relative

5488790

Remove unused imports

771e243

Make links relative

bc6ed60

Merge branch 'master' into master

ba2c713

chihuahua added 3 commits August 12, 2017 17:29

Describe problem of channel stuck at 0

fbb7121

.

9d7f756

.

4abe6e7

dsmilkov merged commit 123a37c into tensorflow:master Aug 13, 2017

GreyZzzzzzXh pushed a commit to GreyZzzzzzXh/tfjs-core that referenced this pull request May 13, 2019

Merge pull request tensorflow#15 from GreyZzzzzzXh/mulmat

f2eaaa5

[Mulmat] register flags for matmul

		@@ -0,0 +1,194 @@
		# Predicting Complementary Colors with deeplearn.js

		@@ -0,0 +1,136 @@
		<!-- Copyright 2017 Google Inc. All Rights Reserved.

		@@ -0,0 +1,355 @@
		/* Copyright 2017 Google Inc. All Rights Reserved.

Write a tutorial for deeplearn.js #15

Write a tutorial for deeplearn.js #15

Uh oh!

Conversation

chihuahua commented Aug 7, 2017 • edited by dsmilkov Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dsmilkov Aug 8, 2017

Choose a reason for hiding this comment

Uh oh!

chihuahua Aug 9, 2017

Choose a reason for hiding this comment

Uh oh!

dsmilkov Aug 8, 2017

Choose a reason for hiding this comment

Uh oh!

chihuahua Aug 9, 2017

Choose a reason for hiding this comment

Uh oh!

dsmilkov Aug 8, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chihuahua Aug 9, 2017

Choose a reason for hiding this comment

Uh oh!

dsmilkov commented Aug 11, 2017

Uh oh!

chihuahua commented Aug 12, 2017

Uh oh!

dsmilkov commented Aug 12, 2017

Uh oh!

chihuahua commented Aug 13, 2017

Uh oh!

chihuahua commented Aug 13, 2017

Uh oh!

chihuahua commented Aug 13, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dsmilkov commented Aug 13, 2017

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chihuahua commented Aug 7, 2017 •

edited by dsmilkov

Loading

dsmilkov Aug 8, 2017 •

edited

Loading

chihuahua commented Aug 13, 2017 •

edited

Loading