def test_simple_pipeline_graph():
graph = build_pipeline_graph(input=MatrixContinuousDense()
,output= MatrixContinuousDense()
,registry=[ExactAlgorithm, HigherInputAlgorithm, LowerOutputAlgorithm]).graph
assert_graph(graph, 3, 3, 6)
graph = build_pipeline_graph(input=List(Text())
,output= Document()
,registry=[WordToWordAlgorithm,
TextToWordAlgorithm,
WordToWordListAlgorithm,
WordListToSentenceAlgorithm,
WordListToSentenceAlgorithm,
SentenceListToDocumentAlgorithm,
TextListToDocumentAlgorithm]).graph
assert_graph(graph, 2, 2, 12)
graph = build_pipeline_graph(input=List(Word())
,output=Document()
,registry=[WordToWordAlgorithm,
TextToWordAlgorithm,
WordToWordListAlgorithm,
WordListToSentenceAlgorithm,
WordListToSentenceAlgorithm,
SentenceListToDocumentAlgorithm,
TextListToDocumentAlgorithm]).graph
assert_graph(graph, 2, 1, 10)
def run(self, input: Sentence()) -> Tuple(List(Word()), List(Flags())):
tokenized = self.nlp(input)
tokens = []
flags = []
for token in tokenized:
token_flags = {}
if self.extract_lemma:
token_flags["lemma"] = token.lemma_
if self.extract_pos_tag:
token_flags["pos"] = token.pos_
for kv in token.tag_.split("|"):
kv = kv.split("=")
if len(kv) == 2:
token_flags["tag_" + kv[0]] = kv[1]
else:
token_flags["tag_" + kv[0]] = True
if self.extract_dep:
token_flags["dep"] = token.dep_
if self.extract_entity:
token_flags["ent_type"] = token.ent_type_
token_flags["ent_kb_id"] = token.ent_kb_id_
if self.extract_details:
token_flags["is_alpha"] = token.is_alpha
token_flags["is_ascii"] = token.is_ascii
token_flags["is_digit"] = token.is_digit
token_flags["is_lower"] = token.is_lower
token_flags["is_upper"] = token.is_upper
token_flags["is_title"] = token.is_title
token_flags["is_punct"] = token.is_punct
token_flags["is_left_punct"] = token.is_left_punct
token_flags["is_right_punct"] = token.is_right_punct
token_flags["is_space"] = token.is_space
token_flags["is_bracket"] = token.is_bracket
token_flags["is_quote"] = token.is_quote
token_flags["is_currency"] = token.is_currency
token_flags["like_url"] = token.like_url
token_flags["like_num"] = token.like_num
token_flags["like_email"] = token.like_email
token_flags["is_oov"] = token.is_oov
token_flags["is_stop"] = token.is_stop
if self.extract_sentiment:
token_flags["sentiment"] = token.sentiment
tokens.append(token.text)
flags.append(token_flags)
return tokens, flags
def test_meta_pipeline_graph():
# Test List algorithm generation
build_pipeline_graph(input=List(Word()),
output=List(Word()),
registry=[WordToWordAlgorithm])
# Test Tuple breakdown feature
build_pipeline_graph(input=Tuple(Word(), Matrix()),
output=Text(),
registry=[WordToWordAlgorithm])
# Test Tuple breakdown feature and List algorithm generation
build_pipeline_graph(input=Tuple(List(Word()), Matrix()),
output=List(Word()),
registry=[WordToWordAlgorithm])
def __init__(
self,
lowercase: Boolean(),
stopwords_remove: Boolean(),
binary: Boolean(),
inner_tokenizer: algorithm(Sentence(), List(Word())),
inner_stemmer: algorithm(Word(), Stem()),
inner_stopwords: algorithm(List(Word()), List(Word())),
):
self.stopwords_remove = stopwords_remove
self.inner_tokenizer = inner_tokenizer
self.inner_stemmer = inner_stemmer
self.inner_stopwords = inner_stopwords
SklearnTransformer.__init__(self)
_CountVectorizer.__init__(self, lowercase=lowercase, binary=binary)
def run(self, input: List(Flags())) -> Flags():
result = {}
for d in input:
result.update(d)
return result
def __init__(
self,
tokenizer: algorithm(Document(), List(Sentence())),
feature_extractor: algorithm(Sentence(), Flags()),
):
self.tokenizer = tokenizer
self.feature_extractor = feature_extractor
def __init__(
self,
tokenizer: algorithm(Sentence(), List(Word())),
feature_extractor: algorithm(Word(), Flags()),
include_text: Boolean(),
):
self.tokenizer = tokenizer
self.feature_extractor = feature_extractor
self.include_text = include_text
def __init__(
self,
extractors: Distinct(
algorithm(Word(), Flags()), exceptions=["MultipleFeatureExtractor"]
),
merger: algorithm(List(Flags()), Flags()),
):
self.extractors = extractors
self.merger = merger
def run(self, input: List(ContinuousVector())) -> ContinuousVector():
input = np.vstack(input)
if self.mode == "mean":
return input.mean(axis=1)
elif self.mode == "max":
return input.max(axis=1)
raise ValueError("Invalid mode: %s" % self.mode)
def _build_pipeline():
builder = build_pipelines(input=Tuple(List(List(Flags())),
List(List(Category()))),
output=List(List(Category())),
registry=find_classes(include="CRF"))
return builder.sample(sampler=Sampler(random_state=0))
def test_save_load_list():
ListClass = build_composite_list(
input_type=Tuple(MatrixContinuousDense(), CategoricalVector()),
output_type=List(CategoricalVector()),
)
algorithm = ListClass(DummyAlgorithm)
fp = BytesIO()
Pickler(fp).dump(algorithm)
fp.seek(0)
algorithm2 = Unpickler(fp).load()
assert repr(algorithm) == repr(algorithm2)
def run(self, input: List(MatrixContinuousDense())) -> Tensor3():
return np.vstack([np.expand_dims(m, axis=0) for m in input])
def run(self, input: List(ContinuousVector())) -> MatrixContinuousDense():
return np.vstack(input)
def run(
self, input: Tuple(Sentence(), List(Tuple(Entity(), Entity(), Category())))
) -> Tuple(List(Vector()), CategoricalVector()):
pass
def __init__(self,
tokenizer: algorithm(Sentence(), List(Word())),
token_feature_extractor: algorithm(Word(), Flags()),
# token_sentence_encoder: algorithm(Word(), )
):
pass
def run(
self, input: Tuple(Sentence(), List(Entity()))
) -> Tuple(List(Word()), List(Postag())):
pass
def run(
self, input: Tuple(List(List(Flags())), List(List(Category())))
) -> List(List(Category())):
return SklearnEstimator.run(self, input)
def run(self, input: List(Sentence())) -> MatrixContinuousSparse():
return SklearnTransformer.run(self, input)
def run(self, input: List(Word())) -> List(Vector()):
pass
def run(
self, input: Tuple(List(MatrixContinuousDense()), List(List(Postag())))
) -> List(List(Postag())):
return super().run(input)
dorothea,
german_credit,
gisette,
shuttle,
wine_quality,
yeast,
]:
X, y, *_ = dataset.load()
run_automl(X, y, name=dataset.__name__)
for dataset in [movie_reviews, haha]:
X, y, *_ = dataset.load()
run_automl(X,
y,
name=dataset.__name__,
input=List(Sentence()),
output=CategoricalVector())
for dataset in [meddocan]:
X, _, y, _ = dataset.load()
run_automl(X,
y,
name=dataset.__name__,
input=List(List(Word())),
output=List(List(Postag())))
for dataset in [cifar10]:
X, y, *_ = dataset.load()
run_automl(X,
y,
name=dataset.__name__,
def run(self, input:List(Text())) -> Document():
pass
parser.add_argument("--token", default=None)
parser.add_argument("--channel", default=None)
args = parser.parse_args()
print(args)
# ## Experimentation
# Instantiate the classifier.
# Note that the input and output types here are defined to match the problem statement,
# i.e., entity recognition.
classifier = AutoML(
search_algorithm=PESearch,
input=List(List(Word())),
output=List(List(Postag())),
search_iterations=args.iterations,
score_metric=meddocan.F1_beta,
cross_validation_steps=1,
search_kwargs=dict(
pop_size=args.popsize,
search_timeout=args.global_timeout,
evaluation_timeout=args.timeout,
memory_limit=args.memory * 1024 ** 3,
),
)
# This custom logger is used for debugging purposes, to be able later to recover
# the best pipelines and all the errors encountered in the experimentation process.
def run(self, input: List(Flags())) -> MatrixContinuousDense():
return super().run(input)
def run(self, input: Sentence()) -> List(Word()):
pass
def run(self, input: Document()) -> List(Flags()):
tokens = self.tokenizer.run(input)
flags = [self.feature_extractor(w) for w in tokens]
return
def run(self, input: List(Vector())) -> Matrix():
pass
def __init__(self, tokenizer: algorithm(Sentence(), List(Word()))):
self.tokenizer = tokenizer
# The next line will print all the algorithms that AutoGOAL found
# in the `contrib` library, i.e., anything that could be potentially used
# to solve an AutoML problem.
for cls in find_classes():
print("Using: %s" % cls.__name__)
# ## Experimentation
# Instantiate the classifier.
# Note that the input and output types here are defined to match the problem statement,
# i.e., text classification.
classifier = AutoML(
search_algorithm=PESearch,
input=List(Sentence()),
output=CategoricalVector(),
search_iterations=args.iterations,
score_metric=f1_score,
search_kwargs=dict(
pop_size=args.popsize,
search_timeout=args.global_timeout,
evaluation_timeout=args.timeout,
memory_limit=args.memory * 1024**3,
),
)
# This custom logger is used for debugging purposes, to be able later to recover
# the best pipelines and all the errors encountered in the experimentation process.
def run(self, input:List(Sentence())) -> Document():
pass