Obtaining underlying representations

Obtaining the representations of documents so you can explore potentially different learning schemes is discussed in the following example:

import autoBOTLib
import pandas as pd

dataframe = pd.read_csv("../data/insults/train.tsv", sep="\t")
train_sequences = dataframe['text_a']
train_targets = dataframe['label']

autoBOTLibObj = autoBOTLib.GAlearner(
    train_sequences,
    train_targets,
    time_constraint=0.1).evolve(representation_step_only = True) ## Construct features only, omit evolution (should be fast).

input_instance_embedding = autoBOTLibObj.transform(train_sequences)

print(input_instance_embedding.shape)

Note that as long as the evolve() was called, the transform() method is able to use the trained vectorizers to obtain sparse (Scipy) matrices.