Building models in NeuNetS

The NeuNetS tool in IBM Watson Studio synthesizes a neural network and trains it on your training data without you having to design or build anything by hand. This topic describes how to use NeuNetS to build a text or image classifier. NeuNetS is currently in beta.

Prerequisite

To use the NeuNetS tool in Watson Studio, you must provision an instance of IBM Watson OpenScale.

See: Watson OpenScale service page in the IBM Cloud catalog external link

Steps

Step 1: Prepare your training data

Note: If you are using sample training data that is available in the NeuNetS tool interface, you can skip this step.

Collect your sample data into the necessary format, depending on the type of data:

Text

  • Create a text file named train.tsv, in which each line contains a sample phrase, a tab character, and a class name

Image

  • Put all training images into a .zip file named train.zip
  • Create a text file named labels.csv, in which each line contains a file name, a comma, and a class name

See:

 

Step 2: Create a project in Watson Studio

  1. In Watson Studio, click the IBM Watson link in the header to navigate to the home panel.
  2. Click New project.
  3. If you are prompted to select a region, choose the US South region.
  4. If you don't already have the required services, such as IBM Cloud Object Storage and IBM Watson Machine Learning, follow the prompts to create new service instances.

 

Step 3: Set up cloud object storage for training data and results

3.1 Navigate to the cloud object storage associated with your Watson Studio project

  1. In Watson Studio, from the Services drop-down menu, right-click "Data Services" and then open Data services in a new browser tab.
  2. In Data services, click on the Cloud Object Storage service.

3.2 Set up training data

Note: If you are using sample training data that is available in the NeuNetS tool interface, you can skip substep 3.2.

  1. Create a bucket for training data. See: Creating a bucket

  2. Upload your training data to the bucket.

    • For text data: upload your train.tsv file
    • For image data: upload both your train.zip file and your labels.csv file

    See: Uploading data to a bucket

3.3 Create a bucket for training results

Create a bucket for training results (the trained model will be saved here, for example.)

See: Creating a bucket

 

Step 4: Open the NeuNetS tool

You can open the NeuNetS tool in two ways:

Option 1

  1. In your Watson Studio project, click Add to project
  2. Click SYNTHESIZED NEURAL NETWORK

Option 2

If you already have a Models section in the Assets tab of your project, click New Synthesized Neural Network.

Note: You will only see NeuNetS options in your project if you have provisioned an instance of IBM Watson OpenScale external link.

 

Step 5: Specify details of your model and training data

  1. Specify a name and description for the new model.
  2. Specify the type of the training data (image, text, or sample.)
  3. If you are using your own training data, follow the prompts to create a connection to your Cloud Object Storage instance, and then specify the training data bucket that you created previously.
  4. Follow the prompts to specify your results bucket.
  5. To prevent IBM from keeping a copy of your training data for internal use in research and product development, check the box.
  6. Click Begin Synthesis.