Custom Model API & Custom Dataset

[saadat-r-ahmed]

I tried to attack a custom sentiment classification model, pysentimiento with a custom dataset, in a csv file.

The model provides a API to make prediction:

from pysentimiento import create_analyzer
analyzer = create_analyzer(task="sentiment", lang="es")

analyzer.predict("Qué gran jugador es Messi")
# returns AnalyzerOutput(output=POS, probas={POS: 0.998, NEG: 0.002, NEU: 0.000})

The dataset is in a .csv format with the columns: "text", "label"

Can help me with the procedure to attack in such a case?
How should I generate a attack report in this case?

[jxmorris12]

You need to write a custom model wrapper, kind of like this one: https://github.com/QData/TextAttack/blob/master/examples/attack/attack_camembert.py