Differential Programming in JavaScript.

FIrst pass, looking for feedback and thoughts on the approach.

Open question: I'm just returning x to add_. Is this sufficient or does x need to be protected?

Let me know your thoughts!

Fixes #264 In Parcel 1.6x, there were some changes made where async functions are transformed to regeneratorRuntime, which is undefined. Used this issue to find a resolution: https://github.com/parcel-bundler/parcel/issues/871

After cloning, running npm install or ./tools/presubmit.js returns an error.

I am running Arch Linux, and my kernel release is 4.12.13-1-ARCH.

Here are the errors:

npm install

/home/adrien/Git/propel/build/Release/libtensorflow_framework.so
  CXX(target) Release/obj.target/tensorflow-binding/src/binding.o
../src/binding.cc: In function ‘napi_value__* HandleGetDevice(napi_env, napi_callback_info)’:
../src/binding.cc:743:50: error: ‘NAPI_AUTO_LENGTH’ was not declared in this scope
   nstatus = napi_create_string_utf8(env, device, NAPI_AUTO_LENGTH, &js_device);
                                                  ^~~~~~~~~~~~~~~~
../src/binding.cc:743:50: note: suggested alternative: ‘NAPI_NO_RETURN’
   nstatus = napi_create_string_utf8(env, device, NAPI_AUTO_LENGTH, &js_device);
                                                  ^~~~~~~~~~~~~~~~
                                                  NAPI_NO_RETURN
../src/binding.cc: In function ‘napi_value__* InitBinding(napi_env, napi_value)’:
../src/binding.cc:1049:7: error: ‘NAPI_AUTO_LENGTH’ was not declared in this scope
       NAPI_AUTO_LENGTH,  // JavasScript class name length
       ^~~~~~~~~~~~~~~~
../src/binding.cc:1049:7: note: suggested alternative: ‘NAPI_NO_RETURN’
       NAPI_AUTO_LENGTH,  // JavasScript class name length
       ^~~~~~~~~~~~~~~~
       NAPI_NO_RETURN
In file included from ../src/binding.cc:15:0:
../src/binding.cc: At global scope:
/home/adrien/.node-gyp/8.5.0/include/node/node_api.h:96:5: error: invalid conversion from ‘napi_value__* (*)(napi_env, napi_value) {aka napi_value__* (*)(napi_env__*, napi_value__*)}’ to ‘napi_addon_register_func {aka void (*)(napi_env__*, napi_value__*, napi_value__*, void*)}’ [-fpermissive]
     };                                                                \
     ^
/home/adrien/.node-gyp/8.5.0/include/node/node_api.h:103:3: note: in expansion of macro ‘NAPI_MODULE_X’
   NAPI_MODULE_X(modname, regfunc, NULL, 0)
   ^~~~~~~~~~~~~
../src/binding.cc:1178:1: note: in expansion of macro ‘NAPI_MODULE’
 NAPI_MODULE(tensorflow_binding, InitBinding)
 ^~~~~~~~~~~
make: *** [tensorflow-binding.target.mk:114: Release/obj.target/tensorflow-binding/src/binding.o] Error 1
make: Leaving directory '/home/adrien/Git/propel/build'

./tools/presubmit.js

Extracting libtensorflow_framework.so
/home/adrien/Git/propel/build/Release/libtensorflow_framework.so
clang -c -o Release/binding.o ../src/binding.cc -I/usr/include/node -I/home/adrien/Git/propel -I/home/adrien/Git/propel/deps/libtensorflow/include -Wall -W -Wno-unused-parameter -std=gnu++0x -DNODE_GYP_MODULE_NAME=tensorflow-binding -DUSING_UV_SHARED=1 -DUSING_V8_SHARED=1 -DV8_DEPRECATION_WARNINGS=1 -D_DARWIN_USE_64_BIT_INODE=1 -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -DBUILDING_NODE_EXTENSION -m64 -fPIC -pthread
../src/binding.cc:743:50: error: use of undeclared identifier 'NAPI_AUTO_LENGTH'
  nstatus = napi_create_string_utf8(env, device, NAPI_AUTO_LENGTH, &js_device);
                                                 ^
../src/binding.cc:1049:7: error: use of undeclared identifier 'NAPI_AUTO_LENGTH'
      NAPI_AUTO_LENGTH,  // JavasScript class name length
      ^
../src/binding.cc:1062:7: error: use of undeclared identifier 'NAPI_AUTO_LENGTH'
      NAPI_AUTO_LENGTH,  // JavasScript class name length
      ^
../src/binding.cc:1178:33: error: cannot initialize a member subobject of type 'napi_addon_register_func' (aka 'void (*)(napi_env__ *,
      napi_value__ *, napi_value__ *, void *)') with an lvalue of type 'napi_value (napi_env, napi_value)' (aka 'napi_value__ *(napi_env__ *,
      napi_value__ *)'): different number of parameters (4 vs 2)
NAPI_MODULE(tensorflow_binding, InitBinding)
                                ^~~~~~~~~~~
/usr/include/node/node_api.h:103:26: note: expanded from macro 'NAPI_MODULE'
  NAPI_MODULE_X(modname, regfunc, NULL, 0)
                         ^~~~~~~
/usr/include/node/node_api.h:92:7: note: expanded from macro 'NAPI_MODULE_X'
      regfunc,                                                        \
      ^~~~~~~

• [x] Created separate profile page per user.
• [x] Added last three notebooks by logged in user to the top of /notebook
  • [x] If user has no notebook, message is displayed (maybe not display anything instead? no header?)
• [x] User's names are clickable and redirecting to user's profile

No error/warning in ./tools/tslint.js This addresses #297

Params wasn’t showing up as gendoc.ts wasnt visiting interfaces (#469 ).

To show the interfaces in the docs I added visitClass in the ts.isInterfaceDeclaration check but to do this I had to change type signatures for visitClass, classElementName and visitProp - not sure if this is a bad way to do things :grimacing: I also added ts.isMethodSignature and ts.isPropertySignature checks to the visitClass function. ... Maybe there’s some nicer way to organise all this?

This caused some interfaces from deps/ to show up (eg Array and Promise) so I filtered out nodes which contain the excludes from tsconfig.json in their declarations file paths.

ConvOpts appears duplicated in the docs as it’s defined in both src/types.ts and src/layers.ts

And I’m not sure why the MinimizeResult interface is showing up in the docs despite not being exported - is this expected?

in_alaska@xxxx propel (grid)⛺  ./tools/presubmit.js 
Delete dir ./build
node ./tools/cpplint.js
python ./deps/cpplint/cpplint.py binding.cc check.h
Done processing binding.cc
Done processing check.h
node ./tools/tslint.js
node ./node_modules/tslint/bin/tslint -p /Users/in_alaska/propel
node ./tools/stylelint.js
node ./node_modules/stylelint/bin/stylelint.js --config stylelint.json website/main.scss website/normalize.scss website/skeleton.scss website/syntax.scss website/syntax_dark.scss website/syntax_light.scss 
node ./tools/test.js
node ./tools/build_binding.js
mkdir /Users/in_alaska/propel/build
mkdir /Users/in_alaska/propel/build/Release
node tools/extract_so.js /Users/in_alaska/propel/build/Release
outDir /Users/in_alaska/propel/build/Release
Extracting /Users/in_alaska/propel/deps/libtensorflow/libtensorflow-cpu-darwin-x86_64.tar.gz
Extracting libtensorflow.so
/Users/in_alaska/propel/build/Release/libtensorflow.so
Extracting libtensorflow_framework.so
/Users/in_alaska/propel/build/Release/libtensorflow_framework.so
clang -c -o Release/binding.o ../binding.cc -I/usr/local/include/node -I/Users/in_alaska/propel/deps/libtensorflow/include -Wall -W -Wno-unused-parameter -std=gnu++0x -DNODE_GYP_MODULE_NAME=tensorflow-binding -DUSING_UV_SHARED=1 -DUSING_V8_SHARED=1 -DV8_DEPRECATION_WARNINGS=1 -D_DARWIN_USE_64_BIT_INODE=1 -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -DBUILDING_NODE_EXTENSION -stdlib=libc++ -mmacosx-version-min=10.7 -arch x86_64 
../binding.cc:703:50: error: use of undeclared identifier 'NAPI_AUTO_LENGTH'
  nstatus = napi_create_string_utf8(env, device, NAPI_AUTO_LENGTH, &js_device);
                                                 ^
../binding.cc:1006:7: error: use of undeclared identifier 'NAPI_AUTO_LENGTH'
      NAPI_AUTO_LENGTH,  // JavasScript class name length
      ^
../binding.cc:1019:7: error: use of undeclared identifier 'NAPI_AUTO_LENGTH'
      NAPI_AUTO_LENGTH,  // JavasScript class name length
      ^
../binding.cc:1123:33: error: cannot initialize a member subobject of type
      'napi_addon_register_func' (aka 'void (*)(napi_env__ *, napi_value__ *, napi_value__
      *, void *)') with an lvalue of type 'napi_value (napi_env, napi_value)' (aka
      'napi_value__ *(napi_env__ *, napi_value__ *)'): different number of parameters
      (4 vs 2)
NAPI_MODULE(tensorflow_binding, InitBinding)
                                ^~~~~~~~~~~
/usr/local/include/node/node_api.h:103:26: note: expanded from macro 'NAPI_MODULE'
  NAPI_MODULE_X(modname, regfunc, NULL, 0)
                         ^~~~~~~
/usr/local/include/node/node_api.h:92:7: note: expanded from macro 'NAPI_MODULE_X'
      regfunc,                                                        \
      ^~~~~~~
4 errors generated.
Error -c
Error ./tools/build_binding.js
Error ./tools/test.js
in_alaska@xxxx propel (grid)⛺  node -v
v8.4.0

• Use byte length, not string length, when computing the number of bytes loaded so far.
• handle http.get() errors.
• Reduce the number of type assertions.
• Explicitly indicate the end of a download by calling downloadProgress(job, null, null).

The deps commit includes some hacks on parcel to fix a bug - I will send upstream soon

https://github.com/propelml/propel_deps/commit/41035369fe03732c3c8eb08cdff617d46dc11caf#diff-5e9e9af54dc82093dc2e5334dfff25c3

Hi How about adding the ability to write a custom description for each published paper just like GitHub's gists. It would be great to have keywords and search for keywords too

As propelml is recommending to use tfjs in it's README, should this repo be archived? Github documentation on Archiving: https://help.github.com/articles/archiving-repositories/

It seems like your build script for dist package is messing with paths.

"[email protected]" from npm contains lots of lines like this one:

var __dirname = "/Users/rld/src/propel/src";

Which causes errors when loading datasets.

If you put { a: "w", b: "hello", c: 123} into the inspector it gives an error: SyntaxError: Unexpected token (4:3)

There's some debate as to whether this is actually a bug, since it's technically a block scope with labels... However the Node REPL and Chrome Inspector parse it in the way one would expect..

Currently the dataset is specified as a string:

pr.dataset("cifar10/train")
pr.dataset("iris")

It would be better to use actual functions for each. example:

pr.dataset.cifar10("train")
pr.dataset.iris()

This allows for type checking on the components of the datasets - that is cifar10 has { images, labels } and iris has { features, labels } - thus the docs will better reflect the information people need.

splitName should default to "train"

cifar10/train is 150mb and mnist/train is 50mb - accessing a single element from either of these will trigger a download of the whole set.

Ideally we can split these datasets into ~ 2mb chunks so they can be downloaded progressively. If only one batch is inspected (like for debugging) only 2mb will be used.

I've already split cifar10/train images into these files and uploaded them (using this script)

http://ar.propelml.org/cifar10_train_images_00.npy
http://ar.propelml.org/cifar10_train_images_01.npy
...
http://ar.propelml.org/cifar10_train_images_49.npy