def _explain(self, X_specimens):
from pyts.bag_of_words import BagOfWords
X_specimens = np.asarray(X_specimens)
size_x = X_specimens.shape[1]
n_model_outputs = self.model.tfidf_.shape[0]
vocab = self.model._tfidf.vocabulary_
normed_tfidf = self.model.tfidf_ / np.linalg.norm(
self.model.tfidf_, axis=1, keepdims=True)
# Matrix of the form:
# 1 -1/3 -1/3 -1/3
# -1/3 1 -1/3 -1/3
# -1/3 -1/3 1 -1/3
# -1/3 -1/3 -1/3 1
mat = np.full((n_model_outputs, n_model_outputs),
-1 / (n_model_outputs - 1))
np.fill_diagonal(mat, 1)
word_impacts = mat @ normed_tfidf
# No numerosity reduction here!
bow = BagOfWords(self.model.window_size,
self.model.window_step,
numerosity_reduction=False)
sentences = bow.transform(self.model._sax.transform(X_specimens))
explanations = []
for x_specimen, sentence in zip(X_specimens, sentences):
sentence = sentence.split(" ")
word_contribs = {
word: word_impacts[:, vocab[word]] /
(cnt if self.divide_spread else 1)
for word, cnt in zip(*np.unique(sentence, return_counts=True))
if word in vocab
}
impacts = np.zeros((n_model_outputs, size_x))
starts_at = 0
for word in sentence:
if word in word_contribs:
impacts[:, starts_at:starts_at + self.model.
window_size] += word_contribs[word][:, np.newaxis]
starts_at += self.model.window_step
explanations.append(TimeExplanation(x_specimen, impacts))
return explanations
:class:`pyts.bag_of_words.BagOfWords`.
"""
# Author: Johann Faouzi <*****@*****.**>
# License: BSD-3-Clause
import matplotlib.pyplot as plt
import numpy as np
from pyts.bag_of_words import BagOfWords
from pyts.datasets import load_gunpoint
# Load the dataset and perform the transformation
X, _, _, _ = load_gunpoint(return_X_y=True)
window_size, word_size = 30, 5
bow = BagOfWords(window_size=window_size,
word_size=word_size,
window_step=window_size,
numerosity_reduction=False)
X_bow = bow.transform(X)
# Plot the considered subseries
plt.figure(figsize=(10, 4))
splits_series = np.linspace(0,
X.shape[1],
1 + X.shape[1] // window_size,
dtype='int64')
for start, end in zip(splits_series[:-1],
np.clip(splits_series[1:] + 1, 0, X.shape[1])):
plt.plot(np.arange(start, end), X[0, start:end], 'o-', lw=1, ms=1)
# Plot the corresponding letters
splits_letters = np.linspace(0, X.shape[1],
import numpy as np
import matplotlib.pyplot as plt
from pyts.bag_of_words import BagOfWords
# Parameters
n_samples, n_timestamps = 100, 48
n_bins = 4
# Toy dataset
rng = np.random.RandomState(42)
alphabet = np.array(['a', 'b', 'c', 'd'])
X_ordinal = rng.randint(n_bins, size=(n_samples, n_timestamps))
X_alphabet = alphabet[X_ordinal]
# Bag-of-words transformation
bow = BagOfWords(window_size=2, numerosity_reduction=False)
X_bow = bow.transform(X_alphabet)
words = np.asarray(X_bow[0].split(' '))
different_words_idx = np.r_[True, words[1:] != words[:-1]]
# Show the results
plt.figure(figsize=(16, 7))
plt.suptitle('Transforming a discretized time series into a bag of words',
fontsize=20,
y=0.9)
plt.subplot(121)
plt.plot(X_ordinal[0], 'o', scalex=0.2)
plt.yticks(np.arange(4), alphabet)
plt.xticks([], [])
plt.yticks(fontsize=16)
def test_actual_results_bag_of_words(params, arr_desired):
"""Test that the actual results are the expected ones."""
arr_actual = BagOfWords(**params).fit_transform(X_bow)
np.testing.assert_array_equal(arr_actual, arr_desired)
def test_parameter_check_bag_of_words(params, error, err_msg):
"""Test parameter validation."""
bow = BagOfWords(**params)
with pytest.raises(error, match=re.escape(err_msg)):
bow.transform(X_bow)