Open yin214 opened 6 months ago
# Basic Information
USER_ID_FIELD: user_id # (str) Field name of user ID feature.
ITEM_ID_FIELD: item_id # (str) Field name of item ID feature.
RATING_FIELD: rating # (str) Field name of rating feature.
TIME_FIELD: timestamp # (str) Field name of timestamp feature.
seq_len: ~ # (dict) Field name of sequence feature: maximum length of each sequence
LABEL_FIELD: label # (str) Expected field name of the generated labels for point-wise dataLoaders.
threshold: ~ # (dict) 0/1 labels will be generated according to the pairs.
NEG_PREFIX: neg_ # (str) Negative sampling prefix for pair-wise dataLoaders.
# Sequential Model Needed
ITEM_LIST_LENGTH_FIELD: item_length # (str) Field name of the feature representing item sequences' length.
LIST_SUFFIX: _list # (str) Suffix of field names which are generated as sequences.
MAX_ITEM_LIST_LENGTH: 50 # (int) Maximum length of each generated sequence.
POSITION_FIELD: position_id # (str) Field name of the generated position sequence.
user_inter_num_interval: "[10,inf)"
item_inter_num_interval: "[10,inf)"
load_col: # (dict) The suffix of atomic files: (list) field names to be loaded.
inter: [user_id, item_id, rating, timestamp]
item: [item_id, categories]
selected_features: [categories]
item_attribute: categories
@yin214 Hello! Thanks for your careful check! We have fixed this bug in #2024
描述这个 bug FEARec模型在sports数据集上训练时会陷入死循环卡住
如何复现 复现这个 bug 的步骤:
hidden_dropout_prob: 0.5 # (float) The probability of an element to be zeroed. attn_dropout_prob: 0.5 # (float) The probability of an attention score to be zeroed.
global_ratio: 0.6 # (float) The ratio of frequency components dual_domain: False # (bool) Frequency domain processing or not std: False # (bool) Use the specific time index or not spatial_ratio: 0.1 # (float) The ratio of the spatial domain and frequency domain fredom: True # (bool) Regularization in the frequency domain or not fredom_type: None # (str) The type of loss in different scenarios topk_factor: 5 # (int) To aggregate time delayed sequences with high autocorrelation
epochs: 100 #训练的最大轮数 train_batch_size: 8192 eval_batch_size: 8192
learning_rate: 0.001
training_neg_sample_num: 1 #负采样数目
eval_step: 1 #每次训练后做evalaution的次数 stopping_step: 10 valid_metric: recall@20
topk: [1,5,10,20]
neg_sampling: ~
eval_args: {'split':{'RS': [0.8,0.1,0.1]}, 'order': 'TO', 'mode': 'full'}