AnchorText- IndexError (dimension mismatch)

Hi, I’m getting this error IndexError: boolean index did not match indexed array along dimension 0; dimension is 100 but corresponding boolean dimension is 1

Any help would be greatly appreciated!!

Here’s the full stack trace.

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-64-b6e19739979d> in <module>
      1 np.random.seed(0)
----> 2 explanation = explainer.explain(text, threshold=0.95, use_unk=True)

/usr/local/lib/python3.6/dist-packages/alibi/explainers/anchor_text.py in explain(self, text, use_unk, use_similarity_proba, sample_proba, top_n, temperature, threshold, delta, tau, batch_size, coverage_samples, beam_size, stop_on_first, max_anchor_size, min_samples_start, n_covered_ex, binary_cache_size, cache_margin, verbose, verbose_every, **kwargs)
    562             verbose=verbose,
    563             verbose_every=verbose_every,
--> 564             **kwargs,
    565         )  # type: Any
    566         result['names'] = [self.words[x] for x in result['feature']]

/usr/local/lib/python3.6/dist-packages/alibi/explainers/anchor_base.py in anchor_beam(self, delta, epsilon, desired_confidence, beam_size, epsilon_stop, min_samples_start, max_anchor_size, stop_on_first, batch_size, coverage_samples, verbose, verbose_every, **kwargs)
    666 
    667         # sample by default 1 or min_samples_start more random value(s)
--> 668         (pos,), (total,) = self.draw_samples([()], min_samples_start)
    669 
    670         # mean = fraction of labels sampled data that equals the label of the instance to be explained, ...

/usr/local/lib/python3.6/dist-packages/alibi/explainers/anchor_base.py in draw_samples(self, anchors, batch_size)
    355         sample_stats, pos, total = [], (), ()  # type: List, Tuple, Tuple
    356         samples_iter = [self.sample_fcn((i, tuple(self.state['t_order'][anchor])), num_samples=batch_size)
--> 357                         for i, anchor in enumerate(anchors)]
    358         for samples, anchor in zip(samples_iter, anchors):
    359             covered_true, covered_false, labels, *additionals, _ = samples

/usr/local/lib/python3.6/dist-packages/alibi/explainers/anchor_base.py in <listcomp>(.0)
    355         sample_stats, pos, total = [], (), ()  # type: List, Tuple, Tuple
    356         samples_iter = [self.sample_fcn((i, tuple(self.state['t_order'][anchor])), num_samples=batch_size)
--> 357                         for i, anchor in enumerate(anchors)]
    358         for samples, anchor in zip(samples_iter, anchors):
    359             covered_true, covered_false, labels, *additionals, _ = samples

/usr/local/lib/python3.6/dist-packages/alibi/explainers/anchor_text.py in sampler(self, anchor, num_samples, compute_labels)
    176         if compute_labels:
    177             labels = self.compare_labels(raw_data)
--> 178             covered_true = raw_data[labels][:self.n_covered_ex]
    179             covered_false = raw_data[np.logical_not(labels)][:self.n_covered_ex]
    180             # coverage set to -1.0 as we can't compute 'true'coverage for this model

IndexError: boolean index did not match indexed array along dimension 0; dimension is 100 but corresponding boolean dimension is 1```

[tiru1930]

@jklaise stub.zip

It needs fasttext , spacy and pandas, alibi

[RobertSamoilescu]

@tiru1930, the main problem is the way you define the prediction function. The internals of Anchor expects the prediction function to be able to predict batches of instances. I believe that in your example your function was expecting a single instance to be classified. Note that the output is expected to be an np.ndarray and it does not matter whether you are predicting the label or the distribution as in the latter case the argmax operator is applied along the axis=1. The second issue is more related to the prediction function of a fasttext model. It seems like the prediction labels are permuted to be in decreasing order of their likelihood. Thus, the most probable class is always placed at the beginning of the prediction output. Taking argmax will always predict label 0. Please check the official documentation to confirm that.

Check the following example

import pandas as pd
import fasttext as ft
from alibi.utils.download import spacy_model
from alibi.explainers import AnchorText
import spacy
import numpy as np

# read dataset and rename category column to label
df = pd.read_csv("test.csv")
df = df.rename(columns={"category": "label"})
df.head()

# add "__label__" prefix for each class label
df["label"] = df["label"].apply(lambda x: "__label__" + x)
df.to_csv("ft_train.csv", sep="\t", index=False, header=False)
print("# of unique labels: %d" % len(df.label.unique()))

# fit the model (you may play a bit with the hyperparams)
model = ft.train_supervised(input="ft_train.csv", lr=1.0, epoch=25, wordNgrams=2)
print(model.test("ft_train.csv"))

# select example to be explained
text = df.text.iloc[100]
label = df.label.iloc[100]
print("Text:", text, "\tLabel:", label)

# load spacy model
smodel = 'en_core_web_md'
spacy_model(model=smodel)
nlp = spacy.load(smodel)

# define the prediction function
predict_fn = lambda x: np.array([model.labels.index(cls[0]) for cls in model.predict(x)[0]])

# define explainer
explainer = AnchorText(
    predictor=predict_fn,
    sampling_strategy='unknown',
    nlp=nlp,
    seed=0
)

# explain the example
explanation = explainer.explain(text, threshold=0.95)

# print explanation
pred = model.labels[predict_fn([text])[0]]
print('Anchor: %s' % (' AND '.join(explanation.anchor)))
print('Precision: %.2f' % explanation.precision)
print('\nExamples where anchor applies and model predicts %s:' % pred)
print('\n'.join([x for x in explanation.raw['examples'][-1]['covered_true']]))