Adding Labels

What is a label?

Classifier bots analyse data from samples and maps this input into the relevant label from among given categories. In other words, the label is the value which is meant to be predicted (output) by the AI neural network, given the sample’s specifications (features).

Classifier bots analyse data from samples and map this input into the relevant label from among given categories. In other words, the label is the value which is meant to be predicted (output) by the AI neural network, given the sample’s specifications (features).

A classifier can have two (binary) or more labels. For instance, for the QSAR biodegradation classifier, the bot predicts whether a molecule is biodegradable (positive label) or not (negative label), based on the molecule’s chemical properties. It also computes the corresponding confidence value to help the user in making the final decision

On the other hand, the bird’s classifier bot is trained to predict birds’ species from among the nine existing labels. The confidence values of the three first most probable species are also computed to help the user choose.

Choosing labels

You can add as many labels as you want when building your prediction bot. Nevertheless, we recommend providing a certain amount of training samples per label to get a meaningful neural network model. Moreover, we recommend keeping the data amount at the same scale for each label to avoid performance misinterpretation.

Adding a new label

The label number is null when you start an AI project on Hailp. To create a new label, you will need first to go to the LABELS tab, and click the “Add new” button. Then you will need to specify the name of your chosen label. The label name must be unique and alphanumeric and must not contain any spaces. If the name contains more than 10 characters, it will be trimmed.

Once a label is created, you can populate it with the corresponding training samples by clicking on the relevant name. From the label upload tab, you can add your entire sample files by simply dragging and dropping. If you selected any data type other than “Single CSV file” during the earlier process, each uploaded file will represent a data instance. In any case, your sample file must have the same structure as the sample you uploaded when creating data features. The following is a typical example of such a file:

cucumber_qualities.csv

  • Color, Width, Length, Weight, QUALITY
  • 1, 50, 100, 20.4, Good
  • 2, 40, 80, 10.4, Bad
  • 1, 10, 120, 26, Good
  • 2, 50, 100, 11.9, Good
  • 1, 60, 150, 20.0, Intermediate
  • 2, 11, 200, 05.5, Bad

Altering a label

You can rename or delete a label as you go. You may also want to empty the entire label before uploading new sample files.

Note that modifying labels, adding new sample files or removing existing files will not affect the final training set, as long as you don’t implement the change. You will need to click on the “Prepare Data” button located in the PREPARE tab to rebuild the training dataset. You will also have to retrain the AI model to implement the change.