Creating a New Image Classifier

To create a new image classifier/model , you need dataset. Image datasets are very large in size so it is recommended to have these image datasets on the Smart Bot in a pre-configured location. So there are three steps in creating a new Image classifier:

• Configuring Dataset Folders on Smart Bot
• Adding Datasets
• Creating/Training a Classifier Model

Configuring Dataset Folders on Smart Bot:

1. Access Smart Bot through WinSCP.
2. Provide IP with the credential file.
3. Navigate to Smartbot_bert > Smartbot_data > nlp_data > datasets.
4. Create Folder for Image Classification dataset.
5. Create the required Datasets folder. 

Each dataset folder contains different folders which represent different classes.  You need to put Images with respect to classes in the respective folders. So all the images from one folder represent one class. 

Adding Datasets in the CC:

1. Navigate to Smart Bot > Classification > Image Classification.

The Image Classification page displays two tabs namely Dataset and Model.

2. Open the Dataset page.
3. Click Add Dataset. 

4. Enter a unique Dataset Name. It should start with a letter and can contain only letters, numbers, space and underscore.
5. Optionally, enter an additional Description.
6. Enter the dataset folder name from the Smart Bot in the Dataset Location field. 

Creating/Training a Classifier Model

Model page helps to select uploaded datasets from Smart Bots from which you can train the model.

1. Navigate to Smart Bot > Classification > Image Classification.

The Image Classification page displays two tabs namely Dataset and Model.

2. Open the Model page.
3. Click Add Model. 

Classifier Configuration window is displayed.

4. Select Create. 
5. Enter the unique Classifier Name. Classifier names cannot start with a number or contain any spaces, and it should contain only alphanumeric letters (a-z) and (0-9), or underscores (_). 
6. Select the required Dataset from the drop-down. 
7. Select Method option from the drop-down. The supported methods are:
   • Convolutional Neural Network (CNN)
   •  Transfer Learning

This selection is optional and are for advanced user. They can change the method to transfer learning.  The default method is CNN. 

8. Optionally, enter an additional Description.
9. Click Submit to initiate a training of the newly created model.

A model is created in the list with the status as ‘In Progress’. The Smart Bot will take some time to train the classifier based on the size of the dataset. When the training is completed, Smart Bot will update the status to ’Completed’.

The Smart Bot also provides the model accuracy in terms of Accuracy Scores, Precision, and Recall.

• Accuracy - Accuracy is the most intuitive performance measure and represents a ratio of correctly predicted observations to the total observations.
• Precision - Precision is the ratio of correctly predicted positive observations to the total predicted positive observations.
• Recall - Recall is the ratio of correctly predicted positive observations to all observations in the actual class.

Precision-Recall is a useful measure of the success of prediction when the classes are very imbalanced. In information retrieval, precision is a measure of result relevancy, while recall is a measure of how many truly relevant results are returned.