def main():
args = docopt.docopt(__doc__)
if args['--data-folder'] is None:
# by default, use 'data' folder relative to this file
args['--data-folder'] = os.path.join(os.path.dirname(__file__), 'data')
if args['add']:
annotate_forms(
data_folder=args['--data-folder'],
url_argument=args["<url>"],
)
elif args['check-data']:
check_annotated_data(args['--data-folder'])
elif args['train']:
ex = FormExtractor.trained_on(
data_folder=args["--data-folder"],
train_ratio=1.0,
)
ex.save(args["<modelfile>"])
elif args['run']:
threshold = float(args['--threshold'])
print("Loading the extractor..")
ex = FormExtractor.load(args["<modelfile>"])
print("Downloading data...")
data, url = load_data(args["<url>"])
tree = load_html(data, url)
result = ex.extract_forms_proba(tree, threshold)
if not result:
print("No forms found.")
return
for form, probs in result:
print("-"*40)
print_form_html(form)
print("")
for tp, prob in Counter(probs).most_common():
tp_full = FORM_TYPES_INV[tp]
print("%s %0.1f%%" % (tp_full, prob*100), end=' ')
print("")
elif args['evaluate']:
n_folds = int(args["--cv"])
ratio = float(args['--test-size'])
store = Storage(args["--data-folder"])
model = get_model()
X, y = store.get_Xy(drop_duplicates=True, verbose=True, leave=True)
test_size = int(len(y) * ratio)
train_size = len(y) - test_size
X_train, X_test, y_train, y_test = X[:train_size], X[train_size:], y[:train_size], y[train_size:]
evaluation.print_metrics(model, X, y, X_train, X_test, y_train, y_test,
ipython=False, cv=n_folds, short_matrix=True)
def annotate_forms(data_folder, url_argument):
"""
Run an interactive HTML form annotation tool.
The process is to download a web page, display all HTML forms and for
each form ask user about form type. The result is saved on disk:
web page is stored as a html file and the URL and the annotation
results are added to index.json file.
"""
storage = Storage(data_folder)
html, url = load_data(url_argument)
doc = load_html(html, url)
answers = _annotate_forms(storage, doc)
if answers:
storage.store_result(html, answers, url)
def annotate_forms(data_folder, url_argument):
"""
Run an interactive HTML form annotation tool.
The process is to download a web page, display all HTML forms and for
each form ask user about form type. The result is saved on disk:
web page is stored as a html file and the URL and the annotation
results are added to index.json file.
"""
storage = Storage(data_folder)
html, url = load_data(url_argument)
doc = load_html(html, url)
answers = _annotate_forms(storage, doc)
if answers:
storage.store_result(html, answers, url)
def main():
args = docopt.docopt(__doc__)
if args['--data-folder'] is None:
# by default, use 'data' folder relative to this file
args['--data-folder'] = os.path.join(os.path.dirname(__file__), 'data')
if args['add']:
annotate_forms(
data_folder=args['--data-folder'],
url_argument=args["<url>"],
)
elif args['check-data']:
check_annotated_data(args['--data-folder'])
elif args['train']:
ex = FormExtractor.trained_on(
data_folder=args["--data-folder"],
train_ratio=1.0,
)
ex.save(args["<modelfile>"])
elif args['run']:
threshold = float(args['--threshold'])
print("Loading the extractor..")
ex = FormExtractor.load(args["<modelfile>"])
print("Downloading data...")
data, url = load_data(args["<url>"])
tree = load_html(data, url)
result = ex.extract_forms_proba(tree, threshold)
if not result:
print("No forms found.")
return
for form, probs in result:
print("-" * 40)
print_form_html(form)
print("")
for tp, prob in Counter(probs).most_common():
tp_full = FORM_TYPES_INV[tp]
print("%s %0.1f%%" % (tp_full, prob * 100), end=' ')
print("")
elif args['evaluate']:
n_folds = int(args["--cv"])
ratio = float(args['--test-size'])
store = Storage(args["--data-folder"])
model = get_model()
X, y = store.get_Xy(drop_duplicates=True, verbose=True, leave=True)
test_size = int(len(y) * ratio)
train_size = len(y) - test_size
X_train, X_test, y_train, y_test = X[:train_size], X[
train_size:], y[:train_size], y[train_size:]
evaluation.print_metrics(model,
X,
y,
X_train,
X_test,
y_train,
y_test,
ipython=False,
cv=n_folds,
short_matrix=True)