https://github.com/masasron/deepnet

DeepNet is simple node based cli tool for creating machine learning classifiers you can use on the web.
https://github.com/masasron/deepnet

classifier classifier-training deep-learning deep-neural-networks machine-learning nodejs

Last synced: about 1 month ago
JSON representation

DeepNet is simple node based cli tool for creating machine learning classifiers you can use on the web.

• Host: GitHub
• URL: https://github.com/masasron/deepnet
• Owner: masasron
• License: mit
• Created: 2018-02-04T05:16:02.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2018-02-11T14:16:20.000Z (almost 7 years ago)
• Last Synced: 2024-10-13T06:51:00.488Z (3 months ago)
• Topics: classifier, classifier-training, deep-learning, deep-neural-networks, machine-learning, nodejs
• Language: JavaScript
• Homepage:
• Size: 43 KB
• Stars: 2
• Watchers: 2
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# DeepNet

DeepNet is simple node based cli tool for creating machine learning classifiers you can use on the web.

```bash

root$ node deepnet/src/cli.js -h

Usage: cli [options] [command]

  Options:

    -V, --version  output the version number

    -h, --help     output usage information

  Commands:

    train [options] 

    make-dataset [options]  

    predict [options]  

```

### Train

```bash

root$ node deepnet/src/cli.js train -h

  Usage: train [options] 

  Options:

    -t, --test-dataset-percentage   percentage of datasets to keep for testing (default: 25)

    -n, --name                  choose a name for your model (default: model-1518027472621)

    -s, --save-period               save model every  iterations (default: 20000)

    -v, --vectorize                 automatically vectorize strings from training data (default: true)

    -l, --learning-rate             network learning rate (default: 0.1)

    -e, --error-threshold           minimum error threshold (default: 0.005)

    -y, --hidden-layers             number of hidden layers (default: 6)

    -i, --iterations                maximum number of iterations (default: 20000)

    -p, --log-period                log progress every  iterations (default: 25)

    -g, --log                       log traning progress (default: true)

    -r, --randomize                 randomize dataset (default: true)

    -a, --activation       activation function (default: sigmoid)

    -h, --help                         output usage information

```


The train command require a JSON dataset file in the format below.   

You may use the `make-dataset` helper command to generate this file.

```json

[

  {

    "input": [0.1,0.2,0.3],

    "output": [0.6]

  },

  {

    "input": [0.1,0,0],

    "output": [0.1]

  }

]

```

### Make Dataset

```sh

root$ node deepnet/src/cli.js make-dataset -h

  Usage: make-dataset [options]  

  Options:

    -n, --name        choose a dataset name (default: dataset-1518028193094)

    -v, --vectorize   automatically vectorize strings (default: true)

    -h, --help           output usage information

```

### Predict

You can use the predict command to load an existing model.

```sh

root$ node deepnetsrc/cli.js predict -h

  Usage: predict [options]  

  Options:

          path to the model .bin file

      test data as string (if -, read from stdin)

    -h, --help   output usage information

```