npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2024 – Pkg Stats / Ryan Hefner

deeplearn-legacy-loader

v0.0.1

Published

Legacy checkpoint loader for deeplearn.js

Downloads

3

Readme

deeplearn.js legacy loader

This repo provides the legacy loader for porting a TensorFlow model to deeplearn.js. Note that this loader is deprecated.

Usage

import {CheckpointLoader} from 'deeplearn-legacy-loader';
const reader = new CheckpointLoader(pathToCheckpointDir);
reader.getAllVariables().then(vars => {
  // vars maps a variable name to a Tensor.
});

Tutorial (Porting MNIST)

All the necessary resources used in this tutorial are stored in the demo/ directory.

To demonstrate the porting steps, we will use a fully connected neural network that predicts hand-written digits from the MNIST dataset. The code is forked from the official TensorFlow MNIST tutorial.

Before we start, make sure you have TensorFlow installed.

First, we clone this repository. We cd into the base dir and train the model in TensorFlow by running:

python demo/fully_connected_feed.py

The training should take ~1 minute and will store a model checkpoint in /tmp/tensorflow/mnist/tensorflow/mnist/logs/fully_connected_feed/.

Next, we need to port the weights from the TensorFlow checkpoint to a format the loader undestands. We provide a script that does this. We run it from the base directory:

python python/dump_checkpoint_vars.py \
  --model_type=tensorflow \
  --output_dir=demo/ \
  --checkpoint_file=/tmp/tensorflow/mnist/logs/fully_connected_feed/model.ckpt-1999

The script will save a set of files (one file per variable, and a manifest.json) in the demo/ directory. The manifest.json is a simple dictionary that maps variable names to files and their shapes:

{
  ...,
  "hidden1/weights": {
    "filename": "hidden1_weights",
    "shape": [784, 128]
  },
  ...
}

To read the weights, we need to create a CheckpointLoader and point it to the manifest file. We then call loader.getAllVariables() which returns a dictionary that maps variable names to Tensors. At that point, we are ready to write our model. Here is a snippet demonstrating the use of CheckpointLoader:

import * as dl from 'deelearn';
import {CheckpointLoader} from 'deeplearn-legacy-loader';

// manifest.json is in the same dir as index.html.
const varLoader = new CheckpointLoader('.');

varLoader.getAllVariables().then(vars => {
  // Get Tensor of variables casted with expected dimension.
  const hidden1W = vars['hidden1/weights'];
  const hidden1B = vars['hidden1/biases'];

  // Write your model here...
});

For details regarding the full model code, see demo/mnist.ts.

To run the mnist demo, run yarn run-demo from the base dir. This compiles the typescript code and runs an http-server on port 8080 that serves the static html/js files.

yarn
yarn run-demo

>> Starting up http-server, serving demo/
>> Available on:
>>   http://127.0.0.1:8080
>>   http://192.168.1.136:8080
>> Hit CTRL-C to stop the server

You should see a simple page showing test accuracy of ~90% measured using a test set of 50 mnist images stored in demo/sample_data.json.