https://github.com/snowch/watson-studio-deep-learning-demo

Steps to demonstrate deep learning in Watson Studio
https://github.com/snowch/watson-studio-deep-learning-demo

deep-learning gpu neural-network watson-studio

Last synced: 3 months ago
JSON representation

Steps to demonstrate deep learning in Watson Studio

• Host: GitHub
• URL: https://github.com/snowch/watson-studio-deep-learning-demo
• Owner: snowch
• License: apache-2.0
• Created: 2018-04-26T15:51:29.000Z (over 7 years ago)
• Default Branch: master
• Last Pushed: 2018-05-07T17:02:39.000Z (over 7 years ago)
• Last Synced: 2025-03-11T03:53:52.198Z (7 months ago)
• Topics: deep-learning, gpu, neural-network, watson-studio
• Language: Jupyter Notebook
• Homepage:
• Size: 40 KB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Watson Studio Deep Learning Demo

### Overview

In this demo you will:

- Create a neural network for classifying hand written digits

- Create an experiment to train the model

- Create a web service to classify

### Pre-requisites

The following IBM Service Instances are required (LITE services are ok):

- Watson Studio

- Watson Machine Learning

- Cloud Object Storage (COS)

   - Create S3 compatible credentials: https://console.bluemix.net/docs/services/cloud-object-storage/iam/service-credentials.html#service-credentials

Ensure that all of the service instances are all in the same IBM Cloud Resource Group, Organisation and Space.

### Watson Studio Steps

This section lists the required Watson Studio setup steps.

#### Project Setup

 - Cloud Object Storage connection

   - Navigate to the Watson Studio Project **Overview** page  

   - Create a connection to your Cloud Object Storage with the name **COS Images**

 - Access Token

   - Navigate to the Watson Studio Project **Settings** page 

   - Create a Project Token with **Editor** Role

 - Import notebooks

   - Import the notebooks **Prepare MNIST Data.ipynb** and **Create Test Image.ipynb**

   - You can use a Python Anaconda Environment (any size will do)

#### Upload MNIST Images

In this section we upload the sample MNIST images to Cloud Object Storage

 - Open the notebook **Prepare MNIST Data**

 - Set the following variables:

   - project_id (this can be found in the URL)

   - project_access_token (this was created in the previous section)

 - Run all cells in the notebook

 - Make a note of the COS bucket names printed at the end

#### Create Neural Network and Training Definition

 - Navigate to the Watson Studio Project **Assets** page

 - Create a new Modeler flow

    - Select **From Example**

    - Select **NEURAL NETWORK DESIGN | Single Convolution layer on MNIST**

    - Click Create

 - Right click the **Image Data** node

 - Select **Open**

    - Under *Data Connections* select **COS Images**

    - Under *Buckets* select [[project-id]]-training

    - Under *Train data file* select **mnist-tf-train.pkl** 

    - Under *Test data file* select **mnist-tf-test.pkl**

    - Under *Validation data file* select **mnist-tf-valid.pkl**

    - Click **Save**

 - On the toolbar, click: **Publish Training Definition**

    - Accept the defaults

    - Click **Publish**

 - In the pop up message click the link **train it in an experiment**

    - Experiment Name: **MNIST Experiment**

    - Under the heading *Cloud Object Storage bucket for storing training source and results files*

       - Click **select**

       - On *Select Cloud Object Storage connection* select **COS Images**

       - Select existing bucket for training data - use training bucket name output from notebook: **Prepare MNIST Data**

       - Select existing bucket for results data - use results bucket name output from notebook: **Prepare MNIST Data**

       - Click **Select**

    - Click **Add training definition**

       - Select **Existing training definition**

       - Under *Existing training definitions* select the training definition exported from the neural network editor

       - Under *Compute plan* select **1/2 x NVIDIA® Tesla® K80 (1 GPU)**

       - Under *Hyperparameter optimization method* select **none**

       - Click **Select**

       - Click **Create and run**

       - When the training run has completed goto the section **Deploy Model**

       

#### Deploy Model

  

   - Click **Actions** on the right of the table for the completed training run **Single Convolution layer on MNIST**

   - Click **Save model**

   - Provide a model name: **MNIST Model**

   - Click **Save**

      - In the message *Model successfully saved. View model details here.* click **here**

      - Click **Deployments**

      - Click **Add Deployment**

         - Provide name **MNIST Deployment**

         - Click **Save**

            - When status is: **DEPLOY_SUCCESS**

            - Select **MNIST Service**

            - Click **Test**

            - Click **Provide Input Data as Json**

            

#### Test deployment

   

   - Open a new Watson Studio browser tab

   - Open the notebook **Create Test Image**

   - Run all cells

   - Copy the output: **{"values": [...]}**

   - Go back the *Deployment Test* tab

       - In the box *Enter input data* paste the copied output

       - Click **Predict*

       

  Note the output probabilities.