def main():
all_data = read_all(directory='../data/03_powertransform')
dims = get_dims(all_data)
dump(dims, '../data/07_dims/dims03.joblib')
all_data = read_all(directory='../data/05_onehot')
dims = get_dims(all_data)
dump(dims, '../data/07_dims/dims05.joblib')
def main():
args = parse_args()
seed_everything(args.seed)
if args.onehot:
all_data = read_all(directory='../data/05_onehot')
sequences = read_sequences(directory='../data/06_onehot_seq')
else:
all_data = read_all(directory='../data/03_powertransform')
sequences = read_sequences(directory='../data/04_sequence')
dims = get_dims(all_data)
loader_maker = LoaderMaker(all_data, sequences, args, onehot=args.onehot)
# CV
name = '15_cnn-onehot' if args.onehot else '15_cnn-label'
skf = StratifiedKFold(n_splits=5)
folds = skf.split(all_data['application_train']['SK_ID_CURR'],
all_data['application_train']['TARGET'])
best_models = []
for train_index, val_index in folds:
train_dataloader = loader_maker.make(train_index)
val_dataloader = loader_maker.make(val_index)
model = LightningModel(R2NCNN(dims, args.n_hidden, args.n_main),
nn.BCEWithLogitsLoss(), train_dataloader,
val_dataloader, args)
trainer = HomeCreditTrainer(name, args.n_epochs, args.patience)
trainer.fit(model)
best_model = load_model(model, name, trainer.logger.version)
best_models.append(best_model)
# Predict
test_dataloader = loader_maker.make(index=None, train=False)
df_submission = predict(best_models, test_dataloader)
filename = '../submission/15_r2n-cnn-onehot.csv' if args.onehot else '../submission/15_r2n-cnn-label.csv'
df_submission.to_csv(filename, index=False)
def main():
args = parse_args()
feature_matrix = joblib.load('../data/02_featuretools/feature_matrix.joblib')
app_train = feature_matrix.dropna(subset=['TARGET'])
app_test = feature_matrix[feature_matrix['TARGET'].isnull()].drop('TARGET', axis=1)
features = read_all('../data/21_dimlstm')
for feature in features.values():
app_train = app_train.merge(feature)
app_test = app_test.merge(feature)
features = read_all('../data/22_vaelstm')
for feature in features.values():
app_train = app_train.merge(feature)
app_test = app_test.merge(feature)
run_lgb(args, app_train, app_test, '36_ftdimvae')
def main():
args = parse_args()
app_train = joblib.load(
'../data/01_labelencoding/application_train.joblib')
app_test = joblib.load('../data/01_labelencoding/application_test.joblib')
features = read_all('../data/21_dimlstm')
for feature in features.values():
app_train = app_train.merge(feature)
app_test = app_test.merge(feature)
features = read_all('../data/22_vaelstm')
for feature in features.values():
app_train = app_train.merge(feature)
app_test = app_test.merge(feature)
run_lgb(args, app_train, app_test, '35_dimvae')
def main():
args = parse_args()
seed_everything(args.seed)
app_train = joblib.load('../data/05_onehot/application_train.joblib')
app_test = joblib.load('../data/05_onehot/application_test.joblib')
sequences = read_all('../data/06_onehot_seq/')
dims = joblib.load('../data/07_dims/dims05.joblib')
app_dims = {}
app_dims['application_train'] = dims.pop('application_train')
app_dims['application_test'] = dims.pop('application_test')
app_data = {'application_train': app_train, 'application_test': app_test}
loader_maker = LoaderMaker(app_data, sequences, args, onehot=True)
skf = StratifiedKFold(n_splits=5)
folds = skf.split(app_train['SK_ID_CURR'], app_train['TARGET'])
best_models = []
for train_index, val_index in folds:
encoders = pretrain(app_train, app_test, sequences, dims, train_index,
val_index, args)
train_dataloader = loader_maker.make(train_index)
val_dataloader = loader_maker.make(val_index)
model = LightningModel(
PretrainedR2N(app_dims, args.n_hidden, args.n_main, encoders),
nn.BCEWithLogitsLoss(), train_dataloader, val_dataloader, args)
name = '82_vaelstm_fine'
trainer = HomeCreditTrainer(name, args.n_epochs, args.patience)
trainer.fit(model)
best_model = load_model(model, name, trainer.logger.version)
best_models.append(best_model)
# Predict
test_dataloader = loader_maker.make(index=None, train=False)
df_submission = predict(best_models, test_dataloader)
df_submission.to_csv(f'../submission/{name}.csv', index=False)
def main():
args = parse_args()
app_train = joblib.load('../data/05_onehot/application_train.joblib')
app_test = joblib.load('../data/05_onehot/application_test.joblib')
sequences = read_all('../data/06_onehot_seq')
dims = joblib.load('../data/07_dims/dims05.joblib')
app_dims = {}
app_dims['application_train'] = dims.pop('application_train')
app_dims['application_test'] = dims.pop('application_test')
encoders = {}
for name, diminfo in dims.items():
model = VAELSTMModule(diminfo, args.n_hidden, None, None, args)
model = load_model(model, name, logdir='../logs/22_vaelstm')
encoder = model.model.encoder
encoders[name] = encoder
run_fine_tuning(args,
app_dims,
app_train,
app_test,
sequences,
encoders,
'42_vaelstm',
onehot=True)
def main():
formatter = '%(asctime)s %(message)s'
logging.basicConfig(filename='../logs/02_featuretools.log', level=logging.INFO, format=formatter)
datas = read_all()
app_train = datas['application_train']
app_test = datas['application_test']
bureau = datas['bureau']
bureau_balance = datas['bureau_balance']
cash = datas['POS_CASH_balance']
previous = datas['previous_application']
installments = datas['installments_payments']
credit = datas['credit_card_balance']
app_test["TARGET"] = np.nan
app = app_train.append(app_test, ignore_index=True, sort=False)
# Entity set with id applications
entity_set = ft.EntitySet(id='HomeCredit')
# Entities with a unique index
entity_set = entity_set.entity_from_dataframe(entity_id='app', dataframe=app, index='SK_ID_CURR')
entity_set = entity_set.entity_from_dataframe(entity_id='bureau', dataframe=bureau, index='SK_ID_BUREAU')
entity_set = entity_set.entity_from_dataframe(entity_id='previous', dataframe=previous, index='SK_ID_PREV')
# Entities that do not have a unique index
entity_set = entity_set.entity_from_dataframe(
entity_id='bureau_balance', dataframe=bureau_balance, make_index=True, index='bureaubalance_index'
)
entity_set = entity_set.entity_from_dataframe(
entity_id='cash', dataframe=cash, make_index=True, index='cash_index'
)
entity_set = entity_set.entity_from_dataframe(
entity_id='installments', dataframe=installments, make_index=True, index='installments_index'
)
entity_set = entity_set.entity_from_dataframe(
entity_id='credit', dataframe=credit, make_index=True, index='credit_index'
)
# Add in the defined relationships
entity_set = entity_set.add_relationships([
ft.Relationship(entity_set['app']['SK_ID_CURR'], entity_set['bureau']['SK_ID_CURR']),
ft.Relationship(entity_set['bureau']['SK_ID_BUREAU'], entity_set['bureau_balance']['SK_ID_BUREAU']),
ft.Relationship(entity_set['app']['SK_ID_CURR'], entity_set['previous']['SK_ID_CURR']),
ft.Relationship(entity_set['previous']['SK_ID_PREV'], entity_set['cash']['SK_ID_PREV']),
ft.Relationship(entity_set['previous']['SK_ID_PREV'], entity_set['installments']['SK_ID_PREV']),
ft.Relationship(entity_set['previous']['SK_ID_PREV'], entity_set['credit']['SK_ID_PREV'])
])
agg_primitives = ['sum', 'count', 'min', 'max', 'mean', 'mode']
feature_matrix, _ = ft.dfs(
entityset=entity_set, target_entity='app', agg_primitives=agg_primitives, max_depth=2, features_only=False, verbose=True
)
feature_matrix = feature_matrix.reset_index()
dump(feature_matrix, '../data/02_featuretools/feature_matrix.joblib')
def main():
all_data = read_all(directory='../data/03_powertransform')
app_train = all_data.pop('application_train')
app_test = all_data.pop('application_test')
data = app_train.append(app_test, sort=False)
data = pd.get_dummies(data)
app_train = data.dropna(subset=['TARGET'])
app_test = data[data['TARGET'].isnull()].drop('TARGET', axis=1)
dump(app_train, '../data/05_onehot/application_train.joblib')
dump(app_test, '../data/05_onehot/application_test.joblib')
with Pool(6) as pool:
pool.map(process, list(all_data.items()))
def main():
args = parse_args()
seed_everything(args.seed)
app_train = joblib.load('../data/03_powertransform/application_train.joblib')
app_test = joblib.load('../data/03_powertransform/application_test.joblib')
sequences = read_all('../data/04_sequence/')
dims = joblib.load('../data/07_dims/dims03.joblib')
app_dims = {}
app_dims['application_train'] = dims.pop('application_train')
app_dims['application_test'] = dims.pop('application_test')
mlflow.set_tracking_uri('../logs/mlruns')
mlflow.set_experiment('HomeCredit')
run_name = '91_dimlstm'
params = vars(args)
df_submission = app_test[['SK_ID_CURR']].copy()
skf = StratifiedKFold(n_splits=5)
folds = skf.split(app_train['SK_ID_CURR'], app_train['TARGET'])
for i, (train_index, val_index) in enumerate(folds):
# Train Encoder
encoders = pretrain(app_train, sequences, dims, train_index, val_index, args)
# Train LightGBM Model
app_encoding_train = predict(app_train, encoders, sequences, args)
x = app_encoding_train.drop(['SK_ID_CURR', 'TARGET'], axis=1)
y = app_encoding_train['TARGET']
x_train, y_train = x.iloc[train_index], y.iloc[train_index]
x_valid, y_valid = x.iloc[val_index], y.iloc[val_index]
train_set = lgb.Dataset(x_train, y_train)
valid_set = lgb.Dataset(x_valid, y_valid)
model = lgb.train(params, train_set, valid_sets=[valid_set])
y_pred = model.predict(x_valid)
auc = roc_auc_score(y_valid, y_pred)
with mlflow.start_run(run_name=run_name):
mlflow.log_params(params)
mlflow.log_metric('auc', auc)
# Predict
app_encoding_test = predict(app_test, encoders, sequences, args)
x_test = app_encoding_test.drop('SK_ID_CURR', axis=1)
y_pred = model.predict(x_test)
df_submission[f'pred_{i}'] = y_pred
df_submission = df_submission.set_index('SK_ID_CURR').mean(axis=1).reset_index()
df_submission.columns = ['SK_ID_CURR', 'TARGET']
df_submission.to_csv(f'../submission/{run_name}.csv', index=False)
def main():
args = parse_args()
app_train = joblib.load('../data/03_powertransform/application_train.joblib')
app_test = joblib.load('../data/03_powertransform/application_test.joblib')
sequences = read_all('../data/04_sequence')
dims = joblib.load('../data/07_dims/dims03.joblib')
app_dims = {}
app_dims['application_train'] = dims.pop('application_train')
app_dims['application_test'] = dims.pop('application_test')
encoders = {}
for name, diminfo in dims.items():
model = DIMLSTMModule(diminfo, args.n_hidden, None, None, args)
model = load_model(model, name, logdir='../logs/21_dimlstm')
encoder = model.encoder
encoders[name] = encoder
run_fine_tuning(args, app_dims, app_train, app_test, sequences, encoders, '41_dimlstm')
def main():
all_data = read_all('../data/05_onehot')
app_train = all_data.pop('application_train')
app_test = all_data.pop('application_test')
df_sk_id_curr = pd.concat(
[app_train[['SK_ID_CURR']], app_test[['SK_ID_CURR']]])
# df_sk_id_curr = df_sk_id_curr.head(100)
bureau = all_data['bureau']
bureau_balance = all_data['bureau_balance']
all_data['bureau_balance'] = pd.merge(
bureau[['SK_ID_CURR', 'SK_ID_BUREAU']],
bureau_balance,
on='SK_ID_BUREAU')
id_list = [df_sk_id_curr] * len(all_data)
with Pool(6) as pool:
pool.starmap(process, zip(id_list, list(all_data.items())))
drop_cols.add(c2)
except TypeError:
continue
df.drop(drop_cols, axis=1, inplace=True)
return df
def process(item):
name, df = item
df = power_transform(df)
df = fillna(df)
df = drop_same_columns(df)
dump(df, f'../data/03_powertransform/{name}.joblib')
if __name__ == "__main__":
datas = read_all()
app_train = datas.pop('application_train')
app_test = datas.pop('application_test')
app = app_train.append(app_test, sort=False)
app = power_transform(app)
app = fillna(app)
app = drop_same_columns(app)
app_train = app.dropna(subset=['TARGET'])
app_test = app[app['TARGET'].isnull()].drop('TARGET', axis=1)
dump(app_train, '../data/03_powertransform/application_train.joblib')
dump(app_test, '../data/03_powertransform/application_test.joblib')
with Pool(6) as pool:
pool.map(process, list(datas.items()))
def main():
args = parse_args()
seed_everything(args.seed)
app_train = joblib.load('../data/05_onehot/application_train.joblib')
app_test = joblib.load('../data/05_onehot/application_test.joblib')
sequences = read_all('../data/06_onehot_seq/')
dims = joblib.load('../data/07_dims/dims05.joblib')
dims.pop('application_train')
dims.pop('application_test')
for name, diminfo in dims.items():
sequence = sequences[name]
train_loader = torch.utils.data.DataLoader(
OneHotSequenceDataset(app_train, sequence),
batch_size=args.batch_size,
shuffle=True,
num_workers=6,
worker_init_fn=worker_init_fn)
test_loader = torch.utils.data.DataLoader(
OneHotSequenceDataset(app_test, sequence),
batch_size=args.batch_size,
shuffle=False,
num_workers=6,
worker_init_fn=worker_init_fn)
model = VAELSTMModule(diminfo, args.n_hidden, train_loader,
test_loader, args)
logdir = '../logs/22_vaelstm'
path = pathlib.Path(logdir) / name
if not path.exists():
path.mkdir(parents=True)
logger = TensorBoardLogger(logdir, name=name)
early_stopping = EarlyStopping(patience=args.patience,
monitor='val_loss',
mode='min')
filepath = pathlib.Path(
logdir) / name / f'version_{logger.version}' / 'checkpoints'
model_checkpoint = ModelCheckpoint(str(filepath),
monitor='val_loss',
mode='min')
trainer = pl.Trainer(default_save_path=logdir,
gpus=-1,
max_epochs=args.n_epochs,
early_stop_callback=early_stopping,
logger=logger,
row_log_interval=100,
checkpoint_callback=model_checkpoint)
trainer.fit(model)
best_model = load_model(model,
name,
trainer.logger.version,
logdir=logdir)
train_loader_no_shuffle = torch.utils.data.DataLoader(
OneHotSequenceDataset(app_train, sequence),
batch_size=args.batch_size,
shuffle=False,
num_workers=6,
worker_init_fn=worker_init_fn)
df_train = predict(name, best_model, train_loader_no_shuffle)
df_test = predict(name, best_model, test_loader)
df_encoding = pd.concat([df_train, df_test])
dump(df_encoding, f'../data/22_vaelstm/{name}.joblib')