def extract():
logging.info('Begin extract')
"""
# Use for batch parsing
candidate_file_agg = list()
for root, subdirs, files in os.walk(lib.get_conf('resume_directory')):
folder_files = map(lambda x: os.path.join(root, x), files) # [os.path.join(root, x) for x in files]
candidate_file_agg.extend(folder_files)
"""
candidate_file = list()
candidate_file.append(sys.argv[1])
# Convert list to a pandas DataFrame
observations = pandas.DataFrame(data=candidate_file, columns=['file_path'])
logging.info('Found {} candidate file(s)'.format(len(observations.index)))
# Subset candidate files to supported extensions
observations['extension'] = observations['file_path'].apply(lambda x: os.path.splitext(x)[1])
observations = observations[observations['extension'].isin(lib.AVAILABLE_EXTENSIONS)]
logging.info('Took candidate file(s) with appropriate file format(s). {} file(s) remain'.
format(len(observations.index)))
with open(candidate_file[0], 'r') as cv: # needs to be utf-8 encoded
text = cv.read()
# Attempt to extract text from files
observations['text'] = text # observations['file_path'].apply(text_extract_utf8)
# Archive schema and return
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations
def load(observations, transformation_pipeline, trained_model):
logging.info('Begin load')
# Reference variables
lib.get_temp_dir()
observations_path = os.path.join(lib.get_temp_dir(), 'observations.csv')
logging.info('Saving observations to path: {}'.format(observations_path))
observations.to_csv(observations_path, index=False)
if transformation_pipeline is not None:
transformation_pipeline_path = os.path.join(
lib.get_temp_dir(), 'transformation_pipeline.pkl')
logging.info('Saving transformation_pipeline to path: {}'.format(
transformation_pipeline))
cPickle.dump(transformation_pipeline,
open(transformation_pipeline, 'w+'))
if trained_model is not None:
trained_model_path = os.path.join(lib.get_temp_dir(),
'trained_model.pkl')
logging.info(
'Saving trained_model to path: {}'.format(transformation_pipeline))
cPickle.dump(trained_model, open(trained_model_path, 'w+'))
lib.archive_dataset_schemas('load', locals(), globals())
logging.info('End load')
pass
def transform(observations, nlp):
logging.info('Begin transform')
print("Extracting name, email, phone, GPA, and dates of work experience")
observations = observations.fillna('')
observations['candidate_name'] = observations['text'].apply(lambda x: field_extraction.candidate_name_extractor(x, nlp))
observations['email'] = observations['text'].apply(lambda x: lib.term_match(x, field_extraction.EMAIL_REGEX))
observations['phone'] = observations['text'].apply(lambda x: lib.term_match(x, field_extraction.PHONE_REGEX))
observations['GPA'] = observations['text'].apply(lambda x: field_extraction.gpa_extractor(x))
observations['years_experience'] = observations['Work'].apply(lambda x: field_extraction.years_of_experience(x))
observations['mos_experience'] = field_extraction.months_of_experience(observations['years_experience'])
# observations['work_dates'] = observations['Work'].apply(
# lambda x: field_extraction.spacy_extractor_by_type(str(x).replace('\n', '. '), nlp, 'DATE', 2))
# observations['uni'] = observations['Edu'].apply(
# lambda x: field_extraction.spacy_extractor_by_type(str(x), nlp, 'ORG', 2))
observations = field_extraction.extract_fields(observations) # search for terms in whole resume
# Archive schema and return
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return observations
def extract():
logging.info('Begin extract')
observations = pandas.DataFrame()
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations
def transform(observations):
logging.info('Begin transform')
# Transform newsgroup20 data set
# Newsgroup20: Extract article filename from document path
observations['filename'] = observations['document_path'].apply(lambda x: ntpath.basename(x))
# Newsgroup20: Extract article category from document path
observations['category'] = observations['document_path'].apply(lambda x: ntpath.basename(os.path.dirname(x)))
# Newsgroup20: Extract article text (and strip article headers), from document path
observations['text'] = observations['document_path'].apply(lambda x: lib.strip_header(open(x).readlines()))
# Remove non-ascii characters
observations['text'] = observations['text'].apply(lambda x: x.decode('ascii', errors='ignore'))
# Newsgroup20: Convert text to normalized tokens. Unknown tokens will map to 'UNK'.
observations['tokens'] = observations['text'].apply(simple_preprocess)
# Newsgroup20: Create bigrams
observations['bigrams'] = observations['text'].apply(lambda x: lib.find_ngrams(x, n=2))
# Newsgroup20: Create modeling text
observations['modeling_text_list'] = observations['tokens'] + observations['bigrams']
observations['modeling_text'] = observations['modeling_text_list'].apply(lambda x: ' '.join(x))
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return observations
def extract():
logging.info('Begin extract')
candidate_file_agg = list() # for creating list of resume file paths
for root, subdirs, files in os.walk(lib.get_conf(
'resume_directory')): # gets path to resumes from yaml file
# os.walk(parentdir + '/data/input/example_resumes'): would do the same thing
files = filter(lambda f: f.endswith(('.pdf', '.PDF')),
files) # only read pdfs
folder_files = map(lambda x: os.path.join(root, x), files)
candidate_file_agg.extend(folder_files)
observations = pd.DataFrame(data=candidate_file_agg,
columns=['file_path']) # convert to df
logging.info('Found {} candidate files'.format(len(observations.index)))
observations['extension'] = observations['file_path'].apply(
lambda x: os.path.splitext(x)[1]) # e.g. pdf or doc
observations = observations[observations['extension'].isin(
lib.AVAILABLE_EXTENSIONS)]
logging.info(
'Subset candidate files to extensions w/ available parsers. {} files remain'
.format(len(observations.index)))
observations['text'] = observations['file_path'].apply(
lib.convert_pdf) # get text from .pdf files
# Archive schema and return
lib.archive_dataset_schemas('extract', locals(),
globals()) # saving the schema
logging.info('End extract')
return observations
def model(observations):
logging.info('Begin model')
# Resources
vocabulary = set(itertools.chain.from_iterable(observations['modeling_text_list']))
vectorizer = CountVectorizer(vocabulary=vocabulary)
# Create train, test sets
msk = numpy.random.rand(len(observations)) < 0.8
train = observations[msk]
test = observations[~msk]
# Create X, y vectors
X_train = vectorizer.fit_transform(train['modeling_text']).todense()
y_train = train['category']
X_test = vectorizer.transform(test['modeling_text']).todense()
y_test = test['category']
# Create, train model
nb = GaussianNB()
nb.fit(X_train, y_train)
# Create predictions, using trained model
test['preds'] = nb.predict(X_test)
scores = nb.score(X_test, y_test)
logging.info('Scores: {}'.format(scores))
lib.archive_dataset_schemas('model', locals(), globals())
logging.info('End model')
return observations, vectorizer, nb, test
def extract():
logging.info('Begin extract')
# Reference variables
candidate_file_agg = list()
# Create list of candidate files
for root, subdirs, files in os.walk(lib.get_conf('resume_directory')):
folder_files = map(lambda x: os.path.join(root, x), files)
candidate_file_agg.extend(folder_files)
# Convert list to a pandas DataFrame
observations = pandas.DataFrame(data=candidate_file_agg,
columns=['file_path'])
logging.info('Found {} candidate files'.format(len(observations.index)))
# Subset candidate files to supported extensions
observations['extension'] = observations['file_path'].apply(
lambda x: os.path.splitext(x)[1])
observations = observations[observations['extension'].isin(
lib.AVAILABLE_EXTENSIONS)]
logging.info(
'Subset candidate files to extensions w/ available parsers. {} files remain'
.format(len(observations.index)))
# Attempt to extract text from files
observations['text'] = observations['file_path'].apply(text_extract_utf8)
# Archive schema and return
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations
def transform(observations, nlp):
# TODO Docstring
logging.info('Begin transform')
# Extract candidate name
observations['candidate_name'] = observations['text'].apply(
lambda x: field_extraction.candidate_name_extractor(x, nlp))
# Extract contact fields
observations['email'] = observations['text'].apply(
lambda x: lib.term_match(x, field_extraction.EMAIL_REGEX))
observations['phone'] = observations['text'].apply(
lambda x: lib.term_match(x, field_extraction.PHONE_REGEX))
# Extract university
observations['universities'] = observations['text'].apply(
field_extraction.extract_universities)
# Extract skills
observations['skills'] = observations['text'].apply(
field_extraction.extract_skills)
# Archive schema and return
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return observations, nlp
def transform(observations, nlp):
# TODO Docstring
logging.info('Begin transform')
# Extract candidate name
observations['candidate_name'] = observations['text'].apply(
lambda x: extract_entities.candidate_name_extractor(x, nlp))
#Extract nationality
observations['nationality'] = observations['text'].apply(
lambda x: extract_entities.nationality_extractor(x, nlp))
# Extract contact fields
observations['email'] = observations['text'].apply(
lambda x: lib.term_match(x, extract_entities.EMAIL_REGEX))
observations['phone'] = observations['text'].apply(
lambda x: lib.term_match(x, extract_entities.PHONE_REGEX))
observations['birthdate'] = observations['text'].apply(
lambda x: lib.birthdate_match(x, extract_entities.BIRTHDATE_REGEX))
observations['unit_postcode'] = observations['text'].apply(
lambda x: lib.term_match(x, extract_entities.UNIT_POSTCODE_REGEX))
observations['url'] = observations['text'].apply(
lambda x: lib.term_match(x, extract_entities.URL_REGEX))
# Extract skills
observations = extract_entities.extract_fields(observations)
# Archive schema and return
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return observations, nlp
def load(mapper, large_model):
"""
- Save mapper to pkl file
- Save large model to h5py file
:param mapper: Mapper, to translate pandas dataframe to usable numpy matrix
:type mapper: DataFrameMapper
:param large_model: A trained keras model
:type large_model: keras.Model
:return:
"""
logging.info('Begin load')
# Save mapper to file
cPickle.dump(mapper, open('../data/output/mapper.pkl', 'w+'))
# Save model to file
large_model.save('../data/output/large_model.h5py', 'w+')
# Archive & return
lib.archive_dataset_schemas('load', locals(), globals())
logging.info('End load')
pass
def model(observations, mapper):
logging.info('Begin model')
cat_vars = ['air_store_id']
cont_vars = ['reserve_visitors']
date_vars = ['visit_datetime', 'reserve_datetime']
response_var = 'visitors'
Xs, y, x_inputs, input_nub, output_nub = df_prep.create_model_layers(
observations, mapper, cat_vars, cont_vars, date_vars, response_var)
# Create model
x = input_nub
preds = output_nub(x)
regression_model = Model(x_inputs, preds)
opt = optimizers.Adam()
regression_model.compile(loss=lib.root_mean_squared_log_error,
optimizer=opt)
regression_model.fit(Xs, y, batch_size=2**12, validation_split=.2)
regression_model.save('..data/models/regression.csv')
lib.archive_dataset_schemas('model', locals(), globals())
logging.info('End model')
pass
def extract():
# Extract appropriate model
char_model = load_model(filepath=lib.get_conf('generate_model_path'))
# Extract posts to be completed
observations = pandas.read_csv(lib.get_conf('post_seed_path'))
logging.info('End extract')
lib.archive_dataset_schemas('generate_extract', locals(), globals())
return char_model, observations
def model(observations):
logging.info('Begin model')
mapper = None
transformation_pipeline = None
trained_model = None
lib.archive_dataset_schemas('model', locals(), globals())
logging.info('End model')
return observations, transformation_pipeline, trained_model
def extract():
# TODO Extract
# Extract all posts for given subreddit, going back given number of days
logging.info('Downloading submissions from Reddit')
observations = scrape_subreddit(lib.get_conf('subreddit'),
lib.get_conf('history_num_days'))
logging.info('Found {} submissions'.format(len(observations.index)))
logging.info('End extract')
lib.archive_dataset_schemas('extract', locals(), globals())
return observations
def extract():
logging.info('Begin extract')
reservations = pandas.read_csv('../data/input/air_reserve.csv')
visits = pandas.read_csv('../data/input/air_visit_data.csv')
observations = pandas.merge(reservations, visits)
observations = observations.sample(frac=1.0, replace=False)
observations = observations.head(100000)
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations
def extract():
"""
Extract necessary data / resources from upstream. This method will:
- Validate that newsgroup data set is available, and read in
- Validate that text embeddings are available, and read in
- Validate that text to embedding index lookup is available, and read in
:return: observations, embedding_matrix, word_to_index
:rtype: (pandas.DataFrame, numpy.array, dict)
"""
logging.info('Begin extract')
logging.info('Performing extract for batch: {}, from newgroup_path: {}'
.format(lib.get_batch_name(), lib.get_conf('newsgroup_path')))
# Download resources
# Confirm newsgroup data set is downloaded
resources.download_newsgroup()
# Confirm that embedding is downloaded
resources.download_embedding()
# Extract resources from file system
# Newsgroup20: Get list of all candidate documents
glob_pattern = os.path.join(lib.get_conf('newsgroup_path'), '*', '*')
logging.info('Searching for glob_pattern: {}'.format(glob_pattern))
document_candidates = glob.glob(glob_pattern)
# Newsgroup20: Create observations data set
observations = pandas.DataFrame(document_candidates, columns=['document_path'])
logging.info('Shape of observations data frame created from glob matches: {}'.format(observations.shape))
# Newsgroup20: Re-order rows
observations = observations.sample(frac=1)
# Newsgroup20: Subset number of observations, if it's a test run
if lib.get_conf('test_run'):
logging.info('Reducing file size for test run')
observations = observations.head(100)
logging.info('Test run number of records: {}'.format(len(observations.index)))
# Embedding: Load embedding
embedding_matrix, word_to_index = resources.create_embedding_matrix()
logging.info('word_to_index max index: {}'.format(max(word_to_index.values())))
# Archive schema and return
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations, embedding_matrix, word_to_index
def transform(observations, label_encoder):
logging.info('Begin transform')
# Feature engineering
observations.columns = map(lambda x: '_'.join(x.lower().split()),
observations.columns)
observations['lat'] = observations['location_1'].apply(
lambda x: eval(x)[0])
observations['long'] = observations['location_1'].apply(
lambda x: eval(x)[1])
# TODO Feature engineering
observations['is_manhattan'] = observations['borough'] == 'MANHATTAN'
observations['is_ny_police'] = observations[
'jurisdiction'] == 'N.Y. POLICE DEPT'
# observations['occurence_epoch'] = pandas.to_datetime(observations['occurrence_datetime'], format='%m/%d/%y %I:%M:%S %p')
# print observations['occurence_epoch'][0], type(observations['occurence_epoch'][0])
# observations['compstat_date'] = observations['compstat_year'].astype(str) + '-' + observations['compstat_month'].astype(str) + '-' + \
# observations['compstat_day'].astype(str)
#
# Dummy out response variable
if label_encoder is None:
label_encoder = lib.create_label_encoder(observations['offense'])
observations['response'] = observations['offense'].apply(
lambda x: label_encoder[x])
observations['is_grand_larceny'] = observations['offense'].apply(
lambda x: x == 'GRAND LARCENY')
logging.info('is_grand_larceny value counts: {}'.format(
observations['is_grand_larceny'].value_counts()))
lib.archive_dataset_schemas('transform', locals(), globals())
regressors = [
'occurrence_day', 'occurrence_year', 'compstat_month', 'compstat_day',
'compstat_year', 'lat', 'long'
]
response_var = 'response'
# TODO Normalization should always be based on training set, not just set at hand
for regressor in regressors:
max_value = observations[regressor].max()
min_value = observations[regressor].min()
observations[regressor] = (observations[regressor] -
min_value) / (max_value - min_value)
regressors.extend(['is_manhattan', 'is_ny_police'])
X = observations[regressors].as_matrix().astype(numpy.float32)
y = numpy.array(observations[response_var].tolist()).astype(numpy.float32)
logging.info('End transform')
return observations, X, y, label_encoder
def load(char_model, observations, generated_posts):
logging.info('Begin transform')
# Export observations
observations.to_csv(
path_or_buf=lib.get_conf('generated_observations_path'), index=False)
# Export generated posts
generated_posts.to_csv(path_or_buf=lib.get_conf('generated_posts_path'),
index=False)
logging.info('End load')
lib.archive_dataset_schemas('generate_load', locals(), globals())
pass
def extract():
"""
- Extract data from CSV
:return:
"""
logging.info('Begin extract')
# Read files from CSV
observations = pandas.read_csv('../data/input/titanic.csv')
# Archive & return
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations
def model(x_train, x_test, y_train, y_test):
"""
- Train multiple models, return a trained model
:param x_train:
:param x_test:
:param y_train:
:param y_test:
:return:
"""
logging.info('Begin model')
# Baseline model
baseline_model = models.baseline()
baseline_model.fit(
x_train,
y_train,
epochs=20,
validation_split=.3,
callbacks=[
TensorBoard(log_dir=os.path.expanduser('~/.logs/baseline'))
])
# Small model
intermediate_model = models.small()
intermediate_model.fit(
x_train,
y_train,
epochs=20,
validation_split=.3,
callbacks=[TensorBoard(log_dir=os.path.expanduser('~/.logs/small'))])
# Large
large_model = models.large()
large_model.fit(
x_train,
y_train,
epochs=20,
validation_split=.3,
callbacks=[TensorBoard(log_dir=os.path.expanduser('~/.logs/large'))])
# Archive & return
lib.archive_dataset_schemas('model', locals(), globals())
logging.info('End model')
return large_model
def transform(observations):
"""
- Convert Sex to boolean male indicator
- Create train / test split
- Create SKLearn-Pandas mapper
- Train SKLearn
- Transform train and test data
:param observations:
:type observations: pandas.DataFrame
:return:
"""
logging.info('Begin transform')
# Convert Sex field into boolean male indicator
observations['male'] = observations['Sex'] == 'male'
logging.info('Converted Sex to binary class. Value counts: {}'.format(
observations['male'].value_counts()))
# Split into train / test split
mask = numpy.random.rand(len(observations)) < 0.8
observations_train = observations[mask]
observations_test = observations[~mask]
logging.info('Creating dataframe mapper')
mapper = DataFrameMapper([(['Age'], [Imputer(),
StandardScaler()]),
(['SibSp'], [Imputer(),
StandardScaler()]),
(['Parch'], [Imputer(),
StandardScaler()]),
(['male'], [Imputer(strategy='most_frequent')])])
logging.info('Fitting and transforming training data set')
x_train = mapper.fit_transform(observations_train)
y_train = observations_train['Survived'].values
logging.info('Transforming response data set')
x_test = mapper.transform(observations_test)
y_test = observations_test['Survived'].values
# Archive & return
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return x_train, x_test, y_train, y_test, mapper
def model(observations, X, y, label_encoder):
logging.info('Beginning model')
# Data split, formatting
dummy_X = observations[['lat', 'long']].as_matrix()
dummy_y = observations['is_grand_larceny']
# ZeroR Model
dummy_clf = DummyClassifier(strategy='constant', constant=1)
dummy_clf.fit(dummy_X, dummy_y)
print('Dummy modle accuracy: {}'.format(dummy_clf.score(dummy_X, dummy_y)))
# Keras model
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.2,
random_state=42)
train_test_mask = numpy.random.random(size=len(observations.index))
num_train = sum(train_test_mask < .8)
num_validate = sum(train_test_mask >= .8)
logging.info(
'Proceeding w/ {} train observations, and {} test observations'.format(
num_train, num_validate))
ff_model = models.gen_stupid_ff_network(X.shape[1], y.shape[1])
ff_model.fit(X_train,
y_train,
batch_size=1024,
epochs=4,
validation_data=(X_test, y_test))
# Add predictions to data set
preds = ff_model.predict(X)
ff_model.metrics, ff_model.metrics_names
observations['max_probability'] = map(max, preds)
observations['prediction_index'] = map(lambda x: numpy.argmax(x), preds)
observations['modeling_prediction'] = map(
lambda x: lib.prop_to_label(x, label_encoder), preds)
trained_model = ff_model
logging.info('End model')
lib.archive_dataset_schemas('model', locals(), globals())
return observations, X, y, label_encoder, trained_model
def load(train, test, transformation_pipeline, trained_model):
"""
Load all assets for downstream use
:param train:
:param test:
:param transformation_pipeline:
:param trained_model:
:return:
"""
logging.info('Begin load')
# Serialize train
train_path = os.path.join(lib.get_batch_output_folder(), 'train.csv')
logging.info('Saving train to path: {}'.format(train_path))
train.to_csv(train_path, index=False)
# Serialize test
test_path = os.path.join(lib.get_batch_output_folder(), 'test.csv')
logging.info('Saving test to path: {}'.format(train_path))
test.to_csv(test_path, index=False)
# Serialize transformation_pipeline
if transformation_pipeline is not None:
transformation_pipeline_path = os.path.join(
lib.get_batch_output_folder(), 'transformation_pipeline.pkl')
logging.info('Saving transformation_pipeline to path: {}'.format(
transformation_pipeline_path))
pickle.dump(transformation_pipeline,
open(transformation_pipeline_path, 'wb'))
# Serialize trained_model
if trained_model is not None:
# Serialize trained_model
trained_model_path = os.path.join(lib.get_batch_output_folder(),
'trained_model.pkl')
logging.info(
'Saving trained_model to path: {}'.format(trained_model_path))
pickle.dump(trained_model, open(trained_model_path, 'wb'))
# Capture model results
print(trained_model.cv_results_)
lib.archive_dataset_schemas('load', locals(), globals())
logging.info('End load')
pass
def model(train, test):
"""
Create a pipeline and train a grid searched model
:param train:
:param test:
:return:
"""
logging.info('Begin model')
mapper = DataFrameMapper([
('honorific', [CountVectorizer(vocabulary=lib.HONORIFIC_VOCABULARY)]),
(['pclass'], [Imputer(), StandardScaler()]),
(['male'], [Imputer(), StandardScaler()]),
(['siblings_spouses_aboard'], [Imputer(), StandardScaler()]),
(['parents_children_aboard'], [Imputer(), StandardScaler()]),
(['fare'], [Imputer(), StandardScaler()]),
])
transformation_pipeline = Pipeline([('featureizer', mapper),
('svc', SVC())])
param_grid = {
'svc__gamma': numpy.logspace(-9, 3, 1),
'svc__C': numpy.logspace(-2, 10, 1),
'svc__kernel': ['linear', 'poly', 'rbf', 'sigmoid'],
'svc__degree': range(2, 8)
}
trained_model = GridSearchCV(transformation_pipeline,
param_grid=param_grid,
scoring='accuracy',
cv=2,
n_jobs=-1)
logging.info('Training model')
trained_model.fit(train.copy(), y=train['survived'])
# Set prediction
for data_set in [train, test]:
data_set['pred'] = trained_model.predict(data_set)
lib.archive_dataset_schemas('model', locals(), globals())
logging.info('End model')
return train, test, transformation_pipeline, trained_model
def transform(observations):
"""
Perform light feature transformation, ahead of feature transformation pipeline
:param observations:
:return:
"""
logging.info('Begin transform')
# Convert the gender column to a male or not column
observations['male'] = observations['sex'] == 'male'
# Get the honorific (e.g. `Mr.` from `,Mr. Henry Jr Sutehall`)
observations['honorific'] = observations['name'].apply(lambda x: str(x).split()[0])
train, test = train_test_split(observations, test_size=0.2)
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return train, test
def load(observations, vectorizer, nb, test):
logging.info('Begin load')
logging.info('Writing observations to CSV')
observations.to_csv(os.path.join(lib.get_batch_output_folder(), 'observations.csv'))
logging.info('Writing test observations to CSV ')
test.to_csv(os.path.join(lib.get_batch_output_folder(), 'test.csv'))
logging.info('Writing vectorizer to file')
cPickle.dump(vectorizer, open(os.path.join(lib.get_batch_output_folder(), 'vectorizer.pkl'), 'w+'))
logging.info('Writing model to file')
cPickle.dump(nb, open(os.path.join(lib.get_batch_output_folder(), 'model.pkl'), 'w+'))
lib.archive_dataset_schemas('load', locals(), globals())
logging.info('End load')
pass
def extract():
"""
Extract the data set from upstream
:return:
"""
logging.info('Begin extract')
# Load the data set
observations = lib.load_titanic()
# Subset observation for speedier test iterations
if lib.get_conf('test_run'):
logging.warn('test_run is set to True. Subsetting to a much smaller data set for testing purposes.')
observations = observations.sample(100)
observations = observations.reset_index()
lib.archive_dataset_schemas('extract', locals(), globals())
logging.info('End extract')
return observations
def extract():
# TODO Docstring
logging.info('Begin extract')
# Extract all posts for given subreddit, going back given number of days
logging.info('Downloading submissions from Reddit')
observations = scrape_subreddit(lib.get_conf('subreddit'),
lib.get_conf('history_num_days'))
logging.info('Found {} submissions'.format(len(observations.index)))
# Load embedding matrix
resources.download_embedding()
embedding_matrix, word_to_index = resources.create_embedding_matrix()
logging.info('word_to_index max index: {}'.format(
max(word_to_index.values())))
logging.info('End extract')
lib.archive_dataset_schemas('extract', locals(), globals())
return embedding_matrix, word_to_index, observations
def transform(observations):
logging.info('Begin transform')
cat_vars = ['air_store_id']
cont_vars = ['reserve_visitors', 'visitors']
date_vars = ['visit_datetime', 'reserve_datetime']
# Convert datetime vars
for date_var in date_vars:
logging.info('Converting date_var: {}'.format(date_var))
observations[date_var] = pandas.to_datetime(observations[date_var],
format='%Y-%m-%d %H:%M:%S')
mapper = df_prep.create_mapper(observations,
cat_vars=cat_vars,
cont_vars=cont_vars,
date_vars=date_vars)
lib.archive_dataset_schemas('transform', locals(), globals())
logging.info('End transform')
return observations, mapper