ezpredict 0.0.6

A small package for convenient predictions on fine-tuned huggingface-models

This package is meant to offer a variety of convenient inference functions, particularly for fine-tuned transformers.

Limited to Jochen Hartmann's fine-tuned j-hartmann/emotion-english-distilroberta-base for now.

Install:

1. Command line pip install ezpredict
2. Google Colab !pip install ezpredict

Imports:

pip install numpy
pip install torch
pip install transformers

How to use:

from ezpredict import predict

preds = predict.predict_input(model_name="j-hartmann/emotion-english-distilroberta-base",
                             input=["What a beautiful day!"],
                              return_values=True,
                              print_values=True)

Variables:

1. model_name -> name of model to perform inference on (limited to Jochen Hartmann's fine-tuned j-hartmann/emotion-english-distilroberta-base for now)
2. input -> list of strings to perform predictions on
3. return_values -> True/False, True: returns predictions as list of tuples
4. print_values -> True/False, True: returns verbose outputs

Sample output:

[('What a beautiful day!', {'anger': 0.0013013791, 'disgust': 0.00047031444, 'fear': 0.001256481, 'joy': 0.95772415, 'neutral': 0.005870249, 'sadness': 0.004233605, 'surprise': 0.029143812}), ('My grandfather died today.', {'anger': 0.0012322478, 'disgust': 0.002399753, 'fear': 0.0018630251, 'joy': 0.0021994421, 'neutral': 0.012114782, 'sadness': 0.96717805, 'surprise': 0.013012686})]