def get_customer_history(flag, n=1000):
in_file = getattr(FileNames, '{}_v2'.format(flag))
out_file = getattr(FileNames, '{}_customer_hist_nn_data'.format(flag))
hist_flag = flag
if flag == 'val':
hist_flag = 'tr'
if flag == 'test':
hist_flag = 'train'
hist_file = getattr(FileNames, 'cust_{}_artifact1'.format(hist_flag))
df = load_pickle(in_file)
hist = load_pickle(hist_file)
agg = ListAggregation(date_col=FieldNames.campaign_start_date,
user_col=FieldNames.customer_id,
key_col=FieldNames.item_set,
hist_artifact=hist)
arr = agg.transform(df)
padded_arr = []
for row in arr:
parr = pad_sequences(row,
maxlen=n,
padding='pre',
truncating='pre',
value=0,
dtype='int32')
padded_arr.append(parr)
padded_arr = np.concatenate(padded_arr)
save_npy(out_file, padded_arr)
def load_data(flag="val"):
if flag == "val":
x_tr = load_pickle(FileNames.tr_features_v1)
x_val = load_pickle(FileNames.val_features_v1)
elif flag == "test":
x_tr = load_pickle(FileNames.train_features_v1)
x_val = load_pickle(FileNames.test_features_v1)
return make_x_y(x_tr, x_val, flag=flag)
def map_campign_id(x_tr, x_val, flag="val"):
if flag == "val":
tr = load_pickle(FileNames.tr_v2)
val = load_pickle(FileNames.val_v2)
elif flag == "test":
tr = load_pickle(FileNames.train_v2)
val = load_pickle(FileNames.test_v2)
x_tr["campaign_id"] = tr["campaign_id"].values
x_val["campaign_id"] = val["campaign_id"].values
return x_tr, x_val
def generate_features(flag):
if flag == "test":
tr_artifact_file = FileNames.train_artifact
hist_artifact_files = [
FileNames.cust_train_artifact1,
FileNames.cust_train_artifact2,
FileNames.cust_train_artifact3,
FileNames.cust_train_artifact4,
]
tr_file = FileNames.train_v2
te_file = FileNames.test_v2
tr_save_file = FileNames.train_features_v1
te_save_file = FileNames.test_features_v1
elif flag == "val":
tr_artifact_file = FileNames.tr_artifact
hist_artifact_files = [
FileNames.cust_tr_artifact1,
FileNames.cust_tr_artifact2,
FileNames.cust_tr_artifact3,
FileNames.cust_tr_artifact4,
]
tr_file = FileNames.tr_v2
te_file = FileNames.val_v2
tr_save_file = FileNames.tr_features_v1
te_save_file = FileNames.val_features_v1
else:
print("flag not VALD!")
tr_artifact = load_pickle(tr_artifact_file)
hist_artifacts = [
load_pickle(hist_file) for hist_file in hist_artifact_files
]
columns = get_feature_names(3)
tr_data = load_pickle(tr_file)
te_data = load_pickle(te_file)
all_data = pd.concat([tr_data, te_data])
pipeline = get_feature_pipeline(tr_artifact, hist_artifacts, all_data)
x_tr = pipeline.fit_transform(tr_data)
x_te = pipeline.transform(te_data)
x_tr = pd.DataFrame(x_tr, columns=columns)
x_te = pd.DataFrame(x_te, columns=columns)
x_tr[FieldNames.target] = tr_data[FieldNames.target].values
if flag == "val":
x_te[FieldNames.target] = te_data[FieldNames.target].values
save_pickle(x_tr, tr_save_file)
save_pickle(x_te, te_save_file)
def save_transaction_artifact(flag):
"""Sace artifacts for customer transactions with different conditions."""
if flag == 'test':
inp_file = FileNames.transaction_test_v1
save_file1 = FileNames.cust_train_artifact1
save_file2 = FileNames.cust_train_artifact2
save_file3 = FileNames.cust_train_artifact3
save_file4 = FileNames.cust_train_artifact4
elif flag == 'val':
inp_file = FileNames.transaction_val_v1
save_file1 = FileNames.cust_tr_artifact1
save_file2 = FileNames.cust_tr_artifact2
save_file3 = FileNames.cust_tr_artifact3
save_file4 = FileNames.cust_tr_artifact4
else:
print('flag not VALID!')
transactions = load_pickle(inp_file)
transactions_grp = group_transactions(transactions)
artifact = _get_transaction_artifact(transactions_grp)
save_pickle(artifact, save_file1)
del artifact, transactions_grp
print("Customer artifact 1 done!")
transactions2 = transactions.loc[
np.abs(transactions[FieldNames.coupon_discount]) > 0]
transactions_grp2 = group_transactions(transactions2)
artifact = _get_transaction_artifact(transactions_grp2)
save_pickle(artifact, save_file2)
del transactions2, transactions_grp2, artifact
print("Customer artifact 2 done!")
transactions3 = transactions.loc[
(np.abs(transactions[FieldNames.coupon_discount]) > 0)
& (np.abs(transactions[FieldNames.other_discount]) > 0)]
transactions_grp3 = group_transactions(transactions3)
artifact = _get_transaction_artifact(transactions_grp3)
save_pickle(artifact, save_file3)
del transactions3, transactions_grp3, artifact
print("Customer artifact 3 done!")
transactions4 = transactions.loc[(np.abs(
transactions[FieldNames.coupon_discount]) > np.abs(
transactions[FieldNames.other_discount]))]
transactions_grp4 = group_transactions(transactions4)
artifact = _get_transaction_artifact(transactions_grp4)
save_pickle(artifact, save_file4)
del transactions4, artifact
print("Customer artifact 4 done!")
def save_train_artifact(flag):
"""Create artifact using training data."""
if flag == 'test':
inp_file = FileNames.train_v2
save_file = FileNames.train_artifact
elif flag == 'val':
inp_file = FileNames.tr_v2
save_file = FileNames.tr_artifact
tr = load_pickle(inp_file)
tr_artifact = HistoricalArtifact(
tr,
user_field=FieldNames.customer_id,
date_field=FieldNames.campaign_start_date,
key_fields=[
FieldNames.campaign_id,
FieldNames.coupon_id,
FieldNames.target,
FieldNames.item_category,
],
)
save_pickle(tr_artifact, save_file)
def get_save_coupon_vecs(flag, brty_map, cat_map):
in_file = getattr(FileNames, '{}_v2'.format(flag))
out_file = getattr(FileNames, '{}_coupon_nn_data'.format(flag))
df = load_pickle(in_file)
coupon_vectors = prepare_coupon_data(df, brty_map, cat_map)
save_npy(out_file, coupon_vectors)