def test_statistics_two_components():
'''
Assert that online lda extracts waited statistics on current document.
'''
n_components = 2
np.random.seed(42)
lda = LDA(n_components, number_of_documents=60)
statistics_list = []
for doc in DOC_SET:
word_list = lda.tokenizer(lda.preprocess(lda._get_text(doc)))
lda._update_indexes(word_list=word_list)
word_indexes = [lda.word_to_index[word] for word in word_list]
statistics, _ = lda._compute_statistics_components(
words_indexes_list=word_indexes, )
statistics_list.append(statistics)
lda._update_weights(statistics=statistics)
for index, statistics in enumerate(statistics_list):
for component in range(n_components):
assert np.array_equal(
a1=statistics[component],
a2=REFERENCE_STATISTICS_TWO_COMPONENTS[index][component],
)
def test_five_components():
'''
Assert that components computed are identical to the original version for n dimensions.
'''
np.random.seed(42)
n_components = 5
online_lda = LDA(
n_components=n_components,
number_of_documents=60,
maximum_size_vocabulary=100,
alpha_beta=100,
alpha_theta=0.5,
)
components_list = []
for document in DOC_SET:
components_list.append(online_lda.fit_transform_one(document))
for index, component in enumerate(components_list):
assert np.array_equal(
a1=list(component.values()),
a2=REFERENCE_FIVE_COMPONENTS[index],
)
def test_five_components():
"""
Assert that components computed are identical to the original version for n dimensions.
"""
n_components = 5
lda = LDA(n_components=n_components,
number_of_documents=60,
maximum_size_vocabulary=100,
alpha_beta=100,
alpha_theta=0.5,
seed=42)
components_list = []
for document in DOC_SET:
tokens = {token: 1 for token in document.split(' ')}
components_list.append(lda.fit_transform_one(tokens))
for index, component in enumerate(components_list):
assert np.array_equal(
a1=list(component.values()),
a2=REFERENCE_FIVE_COMPONENTS[index],
)
def test_extraction_words_ids():
"""
Assert that inputs words are splitted.
Assert that indexes are updated and extractable.
"""
lda = LDA(2, number_of_documents=5, seed=42)
word_indexes_list = []
for doc in DOC_SET:
words = doc.split(' ')
lda._update_indexes(word_list=words)
word_indexes_list.append([lda.word_to_index[word] for word in words])
assert word_indexes_list == [
[1, 2],
[1, 3, 4],
[1, 2, 5],
[1, 3],
[1, 2, 6],
]
def test_extraction_words_ids():
'''
Assert that inputs words are splitted.
Assert that indexes are updated and extractable.
'''
np.random.seed(42)
lda = LDA(2, number_of_documents=5)
word_indexes_list = []
for doc in DOC_SET:
words = lda.process_text(doc)
lda._update_indexes(word_list=words)
word_indexes_list.append([lda.word_to_index[word] for word in words])
assert word_indexes_list == [
[1, 2],
[1, 3, 4],
[1, 2, 5],
[1, 3],
[1, 2, 6],
]
def test_statistics_two_components():
"""
Assert that online lda extracts waited statistics on current document.
"""
n_components = 2
lda = LDA(n_components, number_of_documents=60, seed=42)
statistics_list = []
for doc in DOC_SET:
word_list = doc.split(' ')
lda._update_indexes(word_list=word_list)
word_indexes = [lda.word_to_index[word] for word in word_list]
statistics, _ = lda._compute_statistics_components(
words_indexes_list=word_indexes, )
statistics_list.append(statistics)
lda._update_weights(statistics=statistics)
for index, statistics in enumerate(statistics_list):
for component in range(n_components):
assert np.array_equal(
a1=statistics[component],
a2=REFERENCE_STATISTICS_TWO_COMPONENTS[index][component],
)
def test_prunning_vocabulary():
'''
Vocabulary prunning is available to improve accuracy and limit memory usage.
You can perform vocabulary prunning with parameters vocab_prune_interval (int) and
maximum_size_vocabulary (int).
'''
np.random.seed(42)
online_lda = LDA(n_components=2,
number_of_documents=60,
vocab_prune_interval=2,
maximum_size_vocabulary=3)
components_list = []
for document in DOC_SET:
components_list.append(online_lda.fit_transform_one(x=document))
for index, component in enumerate(components_list):
assert np.array_equal(a1=list(component.values()),
a2=REFERENCE_COMPONENTS_WITH_PRUNNING[index])
def test_fit_transform():
"""
Assert that fit_one and transform_one methods returns waited ouput.
"""
lda = LDA(n_components=2,
number_of_documents=60,
vocab_prune_interval=2,
maximum_size_vocabulary=3,
seed=42)
components_list = []
for document in DOC_SET:
tokens = {token: 1 for token in document.split(' ')}
lda = lda.fit_one(x=tokens)
components_list.append(lda.transform_one(x=tokens))
for index, component in enumerate(components_list):
assert np.array_equal(a1=list(component.values()),
a2=REFERENCE_FIT_ONE_PREDICT_ONE[index])
def test_fit_transform():
'''
Assert that fit_one and transform_one methods returns waited ouput.
'''
np.random.seed(42)
online_lda = LDA(
n_components=2,
number_of_documents=60,
vocab_prune_interval=2,
maximum_size_vocabulary=3,
)
components_list = []
for document in DOC_SET:
online_lda = online_lda.fit_one(x=document)
components_list.append(online_lda.transform_one(x=document))
for index, component in enumerate(components_list):
assert np.array_equal(a1=list(component.values()),
a2=REFERENCE_FIT_ONE_PREDICT_ONE[index])
def test_prunning_vocabulary():
"""
Vocabulary prunning is available to improve accuracy and limit memory usage.
You can perform vocabulary prunning with parameters vocab_prune_interval (int) and
maximum_size_vocabulary (int).
"""
lda = LDA(n_components=2,
number_of_documents=60,
vocab_prune_interval=2,
maximum_size_vocabulary=3,
seed=42)
components_list = []
for document in DOC_SET:
tokens = {token: 1 for token in document.split(' ')}
components_list.append(lda.fit_transform_one(tokens))
for index, component in enumerate(components_list):
assert np.array_equal(a1=list(component.values()),
a2=REFERENCE_COMPONENTS_WITH_PRUNNING[index])
def test_statistics_five_components():
'''
Assert that online lda extracts waited statistics on current document.
'''
np.random.seed(42)
n_components = 5
lda = LDA(
n_components=n_components,
number_of_documents=60,
maximum_size_vocabulary=100,
alpha_beta=100,
alpha_theta=0.5,
)
statistics_list = []
for doc in DOC_SET:
word_list = lda.process_text(doc)
lda._update_indexes(word_list=word_list)
word_indexes = [lda.word_to_index[word] for word in word_list]
statistics, _ = lda._compute_statistics_components(
words_indexes_list=word_indexes, )
statistics_list.append(statistics)
lda._update_weights(statistics=statistics)
for index, statistics in enumerate(statistics_list):
for component in range(n_components):
assert np.array_equal(
a1=statistics[component],
a2=REFERENCE_STATISTICS_FIVE_COMPONENTS[index][component],
)