def get_xy_fd():
feature_dim_dict = {
"sparse": [
SingleFeat('user_age', 4),
SingleFeat('user_gender', 2),
SingleFeat('item_id', 4),
SingleFeat('item_gender', 4)
]
} # raw feature:single value feature
# history behavior feature:multi-value value feature
behavior_feature_list = ["item_id", "item_gender"]
# single value feature input
user_age = np.array([1, 2, 3])
user_gender = np.array([0, 1, 0])
item_id = np.array([0, 1, 2])
item_gender = np.array([0, 1, 0])
# multi-value feature input
hist_item_id = np.array([[0, 1, 2, 3], [0, 1, 2, 3], [0, 1, 2, 0]])
hist_item_gender = np.array([[0, 1, 0, 1], [0, 1, 1, 1], [0, 0, 1, 0]])
# valid length of behavior sequence of every sample
hist_length = np.array([4, 4, 3])
feature_dict = {
'user_age': user_age,
'user_gender': user_gender,
'item_id': item_id,
'item_gender': item_gender,
'hist_item_id': hist_item_id,
'hist_item_gender': hist_item_gender,
}
x = [feature_dict[feat.name] for feat in feature_dim_dict["sparse"]] + \
[feature_dict['hist_'+feat]
for feat in behavior_feature_list] + [hist_length]
# Notice the concatenation order: single feature + multi-value feature + length
# Since the length of the historical sequences of different features in DIN are the same(they are all extended from item_id),only one length vector is enough.
y = [1, 0, 1]
return x, y, feature_dim_dict, behavior_feature_list
def test_DCN_invalid(embedding_size=8, cross_num=0, hidden_size=()):
feature_dim_dict = {
'sparse': [
SingleFeat('sparse_1', 2),
SingleFeat('sparse_2', 5),
SingleFeat('sparse_3', 10)
],
'dense': [
SingleFeat('dense_1', 1),
SingleFeat('dense_1', 1),
SingleFeat('dense_1', 1)
]
}
with pytest.raises(ValueError):
_ = DCN(
feature_dim_dict,
embedding_size=embedding_size,
cross_num=cross_num,
hidden_size=hidden_size,
keep_prob=0.5,
)
data[sparse_features] = data[sparse_features].fillna('-1', )
data[dense_features] = data[dense_features].fillna(0, )
target = ['label']
# 1.Label Encoding for sparse features,and do simple Transformation for dense features
for feat in sparse_features:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])
# 2.count #unique features for each sparse field,and record dense feature field name
sparse_feature_list = [
SingleFeat(feat, data[feat].nunique()) for feat in sparse_features
]
dense_feature_list = [SingleFeat(feat, 0) for feat in dense_features]
# 3.generate input data for model
train, test = train_test_split(data, test_size=0.2)
train_model_input = [train[feat.name].values for feat in sparse_feature_list] + \
[train[feat.name].values for feat in dense_feature_list]
test_model_input = [test[feat.name].values for feat in sparse_feature_list] + \
[test[feat.name].values for feat in dense_feature_list]
# 4.Define Model,train,predict and evaluate
model = DeepFM({
"sparse": sparse_feature_list,
"dense": dense_feature_list
def get_input(use_count=False,
use_unique=False,
use_video=False,
use_audio=False,
use_title=False,
ONLINE_FLAG=False,
SAMPLE_FLAG=True,
VALIDATION_FRAC=0.2,
target='finish'):
train_file = 'track2/sample_train.txt' if SAMPLE_FLAG else 'track2/final_track2_train.txt'
test_file = 'track2/sample_test_no_answer.txt' if SAMPLE_FLAG else 'track2/final_track2_test_no_anwser.txt'
video_file = 'track2/sample_video_features.csv' if SAMPLE_FLAG else 'track2/track2_video_features_mms.csv'
face_file = 'track2/sample_face2.csv' if SAMPLE_FLAG else 'track2/face_df2.csv'
audio_file = 'track2/sample_audio_features.csv' if SAMPLE_FLAG else 'track2/track2_audio_features.csv'
title_file = 'track2/sample_title.txt' if SAMPLE_FLAG else 'track2/track2_title.txt'
data = pd.read_csv(train_file,
sep='\t',
names=[
'uid', 'user_city', 'item_id', 'author_id',
'item_city', 'channel', 'finish', 'like',
'music_id', 'did', 'creat_time', 'video_duration'
])
print('training set read completed.')
if ONLINE_FLAG:
test_data = pd.read_csv(test_file,
sep='\t',
names=[
'uid', 'user_city', 'item_id', 'author_id',
'item_city', 'channel', 'finish', 'like',
'music_id', 'did', 'creat_time',
'video_duration'
])
train_size = data.shape[0]
data = data.append(test_data).reset_index(drop=True)
else:
train_size = int(data.shape[0] * (1 - VALIDATION_FRAC))
print('test set read completed.')
sparse_features = [
'uid',
'user_city',
'item_id',
'author_id',
'item_city',
'channel',
'music_id',
'did',
]
dense_features = []
data['video_duration'] = pd.qcut(data['video_duration'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append('video_duration')
data['creat_time'] = data['creat_time'] % (24 * 3600) / 3600
data['creat_time'] = pd.qcut(data['creat_time'],
q=24,
labels=False,
duplicates='drop')
sparse_features.append('creat_time')
if use_count:
data['uid-author_id'] = data['uid'].astype(
str) + '-' + data['author_id'].astype(str)
data['uid-did'] = data['uid'].astype(str) + '-' + data['did'].astype(
str)
data['did-channel'] = data['did'].astype(
str) + '-' + data['channel'].astype(str)
# 计数特征
cols = ['uid', 'did', 'item_id', 'author_id', 'uid-author_id']
for c in cols:
data[c + '_cnt'] = data[c].map(data[c].value_counts())
data[c + '_cnt'] = pd.qcut(data[c + '_cnt'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append(c + '_cnt')
# 均值特征
df = get_expanding_mean(data[:train_size], data[train_size:], [
'uid-author_id', 'uid-did', 'did-channel', 'uid', 'did', 'item_id'
], 'finish')
dense_features += list(df.columns)
data = pd.concat([data, df], axis=1)
if use_unique:
data['uid_icity_nunique'] = data['uid'].map(
data.groupby('uid')['item_city'].nunique())
data['uid_icity_nunique'] = pd.qcut(data['uid_icity_nunique'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append('uid_icity_nunique')
data['uid_item_nunique'] = data['uid'].map(
data.groupby('uid')['item_id'].nunique())
data['uid_item_nunique'] = pd.qcut(data['uid_item_nunique'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append('uid_item_nunique')
data['uid_author_nunique'] = data['uid'].map(
data.groupby('uid')['author_id'].nunique())
data['uid_author_nunique'] = pd.qcut(data['uid_author_nunique'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append('uid_author_nunique')
data['uid_music_nunique'] = data['uid'].map(
data.groupby('uid')['music_id'].nunique())
data['uid_music_nunique'] = pd.qcut(data['uid_music_nunique'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append('uid_music_nunique')
data['item_ucity_nunique'] = data['item_id'].map(
data.groupby('item_id')['user_city'].nunique())
data['item_ucity_nunique'] = pd.qcut(data['item_ucity_nunique'],
q=10,
labels=False,
duplicates='drop')
sparse_features.append('item_ucity_nunique')
data['item_uid_nunique'] = data['item_id'].map(
data.groupby('item_id')['uid'].nunique())
data['item_uid_nunique'] = pd.qcut(data['item_uid_nunique'],
q=30,
labels=False,
duplicates='drop')
sparse_features.append('item_uid_nunique')
data['author_uid_nunique'] = data['author_id'].map(
data.groupby('author_id')['uid'].nunique())
data['author_uid_nunique'] = pd.qcut(data['author_uid_nunique'],
q=20,
labels=False,
duplicates='drop')
sparse_features.append('author_uid_nunique')
print('generate stats feats completed.')
if use_video:
video_feats = pd.read_csv(video_file)
print('video feats read completed.')
data = pd.merge(data, video_feats, how='left', on='item_id')
for i in range(128):
col = 'vd' + str(i)
data[col].fillna(0, inplace=True)
print('merge video feats completed.')
if use_audio:
audio_feats = pd.read_csv(audio_file)
print('audio feats read completed.')
data = pd.merge(data, audio_feats, how='left', on='item_id')
for i in range(128):
col = 'ad' + str(i)
data[col].fillna(0, inplace=True)
print('merge audio feats completed.')
if use_title:
max_len = 47
title_feats = pd.read_json(title_file, lines=True)
print('title feats read completed')
def get_title_len(d):
return sum(d.values())
title_feats['title_len'] = title_feats['title_features'].apply(
get_title_len)
prior = title_feats['title_len'].mean()
dense_features.append('title_len')
title_feats['title_features'] = title_feats['title_features'].apply(
lambda x: list(x.keys()))
data = pd.merge(data, title_feats, how='left', on='item_id')
for row in data.loc[data.title_features.isna(),
'title_features'].index:
data.at[row, 'title_features'] = []
data['title_len'].fillna(prior, inplace=True)
print('merge title feats completed')
data[sparse_features] = data[sparse_features].fillna('-1', )
for feat in sparse_features:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
if len(dense_features) > 0:
mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])
sparse_feature_list = [
SingleFeat(feat, data[feat].nunique()) for feat in sparse_features
]
dense_feature_list = [SingleFeat(feat, 0) for feat in dense_features]
sequence_feature_list = []
if use_title:
sequence_feature_list.append(
VarLenFeat('title', 134545, max_len, 'sum'))
print('data preprocess completed.')
train = data.iloc[:train_size]
test = data.iloc[train_size:]
train_model_input = [train[feat.name].values for feat in sparse_feature_list] + \
[train[feat.name].values for feat in dense_feature_list]
test_model_input = [test[feat.name].values for feat in sparse_feature_list] + \
[test[feat.name].values for feat in dense_feature_list]
if use_title:
train_model_input += [
pad_sequences(train['title_features'],
maxlen=max_len,
padding='post')
]
test_model_input += [
pad_sequences(test['title_features'],
maxlen=max_len,
padding='post')
]
if use_video:
vd_cols = ['vd' + str(i) for i in range(128)]
video_input = data[vd_cols].values
train_model_input += [video_input[:train_size]]
test_model_input += [video_input[train_size:]]
if use_audio:
ad_cols = ['ad' + str(i) for i in range(128)]
audio_input = data[ad_cols].values
train_model_input += [audio_input[:train_size]]
test_model_input += [audio_input[train_size:]]
print('input process completed.')
print(f'use sparse feats: [{",".join(sparse_features)}]')
print(f'use dense feats: [{",".join(dense_features)}]')
train_labels, test_labels = train[target].values, test[target].values
feature_dim_dict = {
"sparse": sparse_feature_list,
"dense": dense_feature_list,
"sequence": sequence_feature_list
}
if ONLINE_FLAG:
return feature_dim_dict, train_model_input, train_labels, test_model_input, test_labels, test_data
return feature_dim_dict, train_model_input, train_labels, test_model_input, test_labels
def model_pool(defaultfilename='./input/final_track1_train.txt', defaulttestfile='./input/final_track1_test_no_anwser.txt',
defaultcolumnname=['uid', 'user_city', 'item_id', 'author_id', 'item_city', 'channel', 'finish', 'like', 'music_id', 'did', 'creat_time', 'video_duration'],
defaulttarget=['finish', 'like'], defaultmodel="AFM", PERCENT=100):
sparse_features=[]
dense_features=[]
target=defaulttarget
#1 train file
data = pd.read_csv(defaultfilename, sep='\t', names=defaultcolumnname, iterator=True)
#1 train file concats
take=[]
loop = True
while loop:
try:
chunk=data.get_chunk(10000000)
chunk=chunk.sample(frac=PERCENT/100., replace=True, random_state=1)
take.append(chunk)
gc.collect()
except StopIteration:
loop=False
print('stop iteration')
data = pd.concat(take, ignore_index=True, copy=False)
train_size = data.shape[0]
print(train_size)
take.clear()
del [chunk,take]
for column in data.columns:
if column in defaulttarget:
continue
if data[column].dtype in [numpy.float_ , numpy.float64]:
dense_features.append(column)
if data[column].dtype in [numpy.int_, numpy.int64]:
sparse_features.append(column)
# sparse_features=list(set(sparse_features))
# dense_features=list(set(dense_features))
#***************normal
#3. Remove na values
data[sparse_features].fillna('-1', inplace=True)
data[dense_features].fillna(0, inplace=True)
#4. Label Encoding for sparse features, and do simple Transformation for dense features
for feat in sparse_features:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
#5. Dense normalize
if dense_features:
mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])
#*****************normal
#6. generate input data for model
sparse_feature_list = [SingleFeat(feat, data[feat].nunique())
for feat in sparse_features]
dense_feature_list = [SingleFeat(feat, 0)
for feat in dense_features]
#****************model
# 6.choose a model
import pkgutil
import mdeepctr.models
# modelnames = [name for _, name, _ in pkgutil.iter_modules(mdeepctr.__path__)]
# modelname = input("choose a model: "+",".join(modelnames)+"\n")
# if not modelname:
modelname=defaultmodel
# 7.build a model
model = getattr(mdeepctr.models, modelname)({"sparse": sparse_feature_list,
"dense": dense_feature_list}, final_activation='sigmoid', output_dim=len(defaulttarget))
# 8. eval predict
def auc(y_true, y_pred):
return tf.py_func(roc_auc_score, (y_true, y_pred), tf.double)
model.compile("adam", loss="binary_crossentropy", loss_weights=loss_weights, metrics=[auc])
train_model_input = [data[feat.name].values for feat in sparse_feature_list] + \
[data[feat.name].values for feat in dense_feature_list]
train_labels = [data[target].values for target in defaulttarget]
my_callbacks = [EarlyStopping(monitor='loss', min_delta=1e-2, patience=1, verbose=1, mode='min')]
history = model.fit(train_model_input, train_labels,
batch_size=2**14, epochs=3, verbose=1, callbacks=my_callbacks)
del [train_model_input, train_labels, data]
# import objgraph
# objgraph.show_refs([data], filename='data-graph.png')
#2 test file
test_data = pd.read_csv(defaulttestfile, sep='\t', names=defaultcolumnname, )
raw_test_data=test_data.copy()
#data = data.append(test_data)
test_size=test_data.shape[0]
print(test_size)
#***************normal
#3. Remove na values
test_data[sparse_features].fillna('-1', inplace=True)
test_data[dense_features].fillna(0, inplace=True)
#4. Label Encoding for sparse features, and do simple Transformation for dense features
for feat in sparse_features:
lbe = LabelEncoder()
test_data[feat] = lbe.fit_transform(test_data[feat])
#5. Dense normalize
if dense_features:
mms = MinMaxScaler(feature_range=(0, 1))
test_data[dense_features] = mms.fit_transform(test_data[dense_features])
#*****************normal
#test = test_data
test_model_input = [test_data[feat.name].values for feat in sparse_feature_list] + \
[test_data[feat.name].values for feat in dense_feature_list]
pred_ans = model.predict(test_model_input, batch_size=2**14)
result = raw_test_data[['uid', 'item_id', 'finish', 'like']].copy()
result.rename(columns={'finish': 'finish_probability',
'like': 'like_probability'}, inplace=True)
result['finish_probability'] = pred_ans[0]
result['like_probability'] = pred_ans[1]
output = "%s-result.csv" % (modelname)
result[['uid', 'item_id', 'finish_probability', 'like_probability']].to_csv(
output, index=None, float_format='%.6f')
return history
sparse_features = ['uid', 'user_city', 'item_id', 'author_id', 'item_city', 'channel',
'music_id', 'did', ]
dense_features = ['video_duration'] # 'creat_time',
data[sparse_features] = data[sparse_features].fillna('-1', )
data[dense_features] = data[dense_features].fillna(0,)
target = ['finish', 'like']
for feat in sparse_features:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])
sparse_feature_list = [SingleFeat(feat, data[feat].nunique())
for feat in sparse_features]
dense_feature_list = [SingleFeat(feat, 0)
for feat in dense_features]
train = data.iloc[:train_size]
test = data.iloc[train_size:]
train_model_input = [train[feat.name].values for feat in sparse_feature_list] + \
[train[feat.name].values for feat in dense_feature_list]
test_model_input = [test[feat.name].values for feat in sparse_feature_list] + \
[test[feat.name].values for feat in dense_feature_list]
train_labels = [train[target[0]].values, train[target[1]].values]
test_labels = [test[target[0]].values, test[target[1]].values]
key2index = {}
genres_list = list(map(split, data['genres'].values))
genres_length = np.array(list(map(len, genres_list)))
max_len = max(genres_length)
# Notice : padding=`post`
genres_list = pad_sequences(
genres_list,
maxlen=max_len,
padding='post',
)
# 2.count #unique features for each sparse field and generate feature config for sequence feature
sparse_feat_list = [
SingleFeat(feat, data[feat].nunique()) for feat in sparse_features
]
sequence_feature = [VarLenFeat(
'genres',
len(key2index) + 1, max_len,
'mean')] # Notice : value 0 is for padding for sequence input feature
# 3.generate input data for model
sparse_input = [data[feat.name].values for feat in sparse_feat_list]
dense_input = []
sequence_input = [genres_list]
model_input = sparse_input + dense_input + \
sequence_input # make sure the order is right
# 4.Define Model,compile and train
model = DeepFM({
def model_pool(defaultfilename='./input/final_track2_train.txt',
defaulttestfile='./input/final_track2_test_no_anwser.txt',
defaultcolumnname=[
'uid', 'user_city', 'item_id', 'author_id', 'item_city',
'channel', 'finish', 'like', 'music_id', 'did',
'creat_time', 'video_duration'
],
defaulttarget=['finish', 'like'],
defaultmodel="AFM",
PERCENT=1):
data = pd.read_csv(defaultfilename,
sep='\t',
names=defaultcolumnname,
iterator=True)
#1 train file
take = []
loop = True
while loop:
try:
chunk = data.get_chunk(10000)
chunk = chunk.take(list(range(min(chunk.shape[0], PERCENT * 100))),
axis=0)
take.append(chunk)
except StopIteration:
loop = False
print('stop iteration')
data = pd.concat(take, ignore_index=True)
train_size = data.shape[0]
print(train_size)
#2 extract file
test_data = pd.read_csv(
defaulttestfile,
sep='\t',
names=defaultcolumnname,
)
data = data.append(test_data)
test_size = test_data.shape[0]
print(test_size)
sparse_features = []
dense_features = []
target = defaulttarget
for column in data.columns:
if column in defaulttarget:
continue
if data[column].dtype in [numpy.float_, numpy.float64]:
dense_features.append(column)
if data[column].dtype in [numpy.int_, numpy.int64]:
sparse_features.append(column)
#3. Remove na values
data[sparse_features] = data[sparse_features].fillna('-1', )
data[dense_features] = data[dense_features].fillna(0, )
#4. Label Encoding for sparse features, and do simple Transformation for dense features
for feat in sparse_features:
lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
#5. Dense normalize
if dense_features:
mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])
#6. generate input data for model
sparse_feature_list = [
SingleFeat(feat, data[feat].nunique()) for feat in sparse_features
]
dense_feature_list = [SingleFeat(feat, 0) for feat in dense_features]
#7. generate input data for model
train = data.iloc[:train_size]
test = data.iloc[train_size:]
#8.generate data
print(train.columns)
train_model_input = [train[feat.name].values for feat in sparse_feature_list] + \
[train[feat.name].values for feat in dense_feature_list]
test_model_input = [test[feat.name].values for feat in sparse_feature_list] + \
[test[feat.name].values for feat in dense_feature_list]
train_labels = [train[target].values for target in defaulttarget]
test_labels = test[target]
# 6.choose a model
import pkgutil
import mdeepctr.models
# modelnames = [name for _, name, _ in pkgutil.iter_modules(mdeepctr.__path__)]
# modelname = input("choose a model: "+",".join(modelnames)+"\n")
# if not modelname:
modelname = defaultmodel
# 7.build a model
model = getattr(mdeepctr.models, modelname)({
"sparse": sparse_feature_list,
"dense": dense_feature_list
},
final_activation='sigmoid',
output_dim=len(defaulttarget))
# 8. eval predict
def auc(y_true, y_pred):
return tf.py_func(roc_auc_score, (y_true, y_pred), tf.double)
# model.compile("adagrad", loss="binary_crossentropy", metrics=[auc])
model.compile(optimizer="adam", loss="binary_crossentropy", metrics=[auc])
my_callbacks = [
EarlyStopping(monitor='loss',
min_delta=1e-2,
patience=3,
verbose=1,
mode='min')
]
history = model.fit(train_model_input,
train_labels,
batch_size=4096,
epochs=100,
verbose=1,
callbacks=my_callbacks)
pred_ans = model.predict(test_model_input, batch_size=2**14)
# nsamples, nx, ny = numpy.asarray(pred_ans).shape
# pred_ans = numpy.asarray(pred_ans).reshape((nx*ny, nsamples))
# print(test_labels.shape)
# print(pred_ans.shape)
#
# logloss = round(log_loss(test_labels, pred_ans), 4)
# try:
# roc_auc = round(roc_auc_score(test_labels, pred_ans), 4)
# except:
# roc_auc=0
result = test_data[['uid', 'item_id', 'finish', 'like']].copy()
result.rename(columns={
'finish': 'finish_probability',
'like': 'like_probability'
},
inplace=True)
result['finish_probability'] = pred_ans[0]
result['like_probability'] = pred_ans[1]
output = "%s-result.csv" % (modelname)
result[['uid', 'item_id', 'finish_probability',
'like_probability']].to_csv(output,
index=None,
float_format='%.6f')
return history
def deepctr_cv(X_train,
y_train,
folds,
logger,
cv_path,
X_test=None,
optional_data=None,
prep=True,
split_conf=None):
scores = []
preds = []
meta = np.zeros_like(y_train).astype("float64")
if split_conf is None:
X_tr, X_te, main_conf, _ = prep_for_embedding(X_train,
X_test,
conf,
prep=prep)
X_train, X_test = X_tr, X_te
else:
main_conf = split_conf
cat_cols = [c for c, _, _ in main_conf[0]]
cat_fs = [SingleFeat(c, d) for c, d, _ in main_conf[0]]
num_fs = [SingleFeat(c, 0) for c in conf.num_cols]
X_test = split_df(X_test, cat_cols, conf.num_cols)
for num_fold, (tr_ind, tes_ind) in enumerate(folds):
if num_fold > 0:
break
logger.info(f"fold_{num_fold}")
fold_path = cv_path / f"fold{num_fold}"
seed_path = fold_path
Path(fold_path).mkdir(exist_ok=True, parents=True)
callbacks = [CSVLogger(str(fold_path / 'epochs.csv'))]
X_cv_train, X_cv_test = X_train.iloc[tr_ind], X_train.iloc[tes_ind]
y_cv_train, y_cv_test = y_train.iloc[tr_ind], y_train.iloc[tes_ind]
X_cv_train = split_df(X_cv_train, cat_cols, conf.num_cols)
X_cv_test = split_df(X_cv_test, cat_cols, conf.num_cols)
model = DeepFM({
'sparse': cat_fs,
'dense': num_fs
},
final_activation='sigmoid')
model.compile("adam", "binary_crossentropy", metrics=['accuracy'])
model.fit(X_cv_train,
y_cv_train,
callbacks=callbacks,
batch_size=2048,
epochs=10,
verbose=1,
validation_data=(X_cv_test, y_cv_test))
model.save_weights(str(seed_path / 'weights.h5'), save_format='hdf5')
gc.collect()
if X_test is not None:
pred = model.predict(X_test, batch_size=2048)
pred = pred[:, 0]
np.save(seed_path / f"pred.npy", pred)
train_oof = model.predict(X_cv_test, batch_size=2048)
train_oof = train_oof[:, 0]
auc = roc_auc_score(y_cv_test.values, train_oof)
logger.info(f"{num_fold}: auc {auc}")
np.save(seed_path / f"train_oof.npy", train_oof)
# auc = roc_auc_score(y_cv_test, train_oof)
# logger.info(f"seed_average: auc {auc}")
scores.append(auc)
np.save(fold_path / f"tes_ind.npy", tes_ind)
meta[tes_ind] += train_oof
del X_cv_train, y_cv_train, X_cv_test, y_cv_test
if X_test is not None:
preds.append(pred)
scores = np.array(scores)
preds = np.array(preds)
pred = rank_average(preds)
logger.info(f"{scores.mean()}, {scores.std()}")
return scores, pred, meta