Zero Shot Text Classifier

The node performs Zero-Shot Text Classification based on NLI task. No prior training is needed, user can just provide a set of labels that are expected to be used for text classification.
The node can use a model trained on an NLI task. It works by posing each candidate label as a 'hypothesis' and the sequence which one wants to classify as the 'premise'.

Options

Settings

Sentence column

A column with a plain text (String), which contains the text to be classified. No special pre-processing is needed.

Candidate labels

A set of labels to use for prediction.
E.g : Positive;Negative

Use custom labels separator

If active, it is possible to change the character used to separate different candidate labels.

Use custom hypothesis

If active, it is possible to change the hypothesis used as an input to the model. Otherwise the default hypothesis will be used.

Batch size

The number of samples that are passed to the model at once for one batch. It highly depends on the RAM or VRAM.

Change prediction column name

If active a column with provided name will be created in the output table. Otherwise the default name will be used for the column with predictions.

Append individual class probabilities

If active, the columns with class probabilities will be created in the output table.

Multi-label

Multi-label classification

If active, it is possible to change the probability threshold for assigning the class or the number of desired classes. By default one class is assigned by the biggest probability value.

Use custom threshold for assigning the classes

If active, it is possible to change the probability threshold for assigning the class. By default class is assigned by the biggest probability value.

Probability threshold

The class is assigned if the class probability is equal or higher then the value. Several classes might be assigned.

Fixed number of classes per prediction

If active, it it possible to change the number of assigned classes per prediction.

Number of classes per prediction

The number of classes that will be assigned to the prediction.

Python

Python

Select one of Python execution environment options:

• use default Python environment for Deep Learning
• use Conda environment

Views

This node has no views

Workflows

• Huminfra Handbook WorkflowKNIME Hub
• Zero Shot Text ClassificationKNIME Hub
• Zero Shot Text Classification with BERTKNIME Hub

Links

• No links available

Developers