isssue of ranking based measures

[Maryam483]

hello in config file when i make ranking:True since i need ranking based evaluation metrics, it gives errors... below is my config file code, please help
field_separator: ","
seq_separator: " "

gpu_id: 0
use_gpu: True
show_progress: False
save_dataset: False
save_dataloaders: False

############### data setting ###############
seed: 2022
dataset: depaulmovie

define data_path as the parent directory of your data folder

data_path: d:\dataset\

USER_ID_FIELD: user_id
ITEM_ID_FIELD: item_id
RATING_FIELD: rating
CONTEXT_SITUATION_FIELD: contexts
USER_CONTEXT_FIELD: uc_id

note: you can use either load or unload, cannot use them both

load_col is used to load specific columns; unload_col is used to ignore selected columns

set "load_col: ~", if you want to load all cols

load_col: {'inter': ['user_id','item_id','rating','contexts','uc_id']}

unload_col: {'inter': ['contexts']}

by default, we load all cols, unless there are some special requirements

load_col: ~
#load_col: {'inter': ['user_id','item_id','rating','contexts','uc_id']} # Add 'time' if it's needed

used for topN ranking only

LABEL_FIELD: label
threshold:
rating: 0

the current library does not support negative sampling

neg_sampling: ~

############### model setting ###############
model: NeuCMFii

General model

epochs: 50
train_batch_size: 5000
eval_batch_size: 409600
learner: adam

learner: adam, RMSprop

stopping_step: 10
clip_grad_norm: ~

clip_grad_norm: {'max_norm': 5, 'norm_type': 2}

weight_decay: 0.0

NeuCF models

mf_embedding_size: 64
mlp_embedding_size: 64
mlp_hidden_size: [128,64,32]
learning_rate: 0.01
dropout_prob: 0.1

#tf_train: True
mf_train: True
mlp_train: True

FM models

embedding_size: 64
#mlp_hidden_size: [128,64,32]
#learning_rate: 0.01
#dropout_prob: 0.3

############### Evaluation setting ###############
eval_args:

split: {'RS': [0.8, 0.2]} # hold-out evaluation

split: {'CV': 5} # N-fold cross validation
group_by: user
mode: labeled # do not change it, DeepCARSKit only support this mode
order: RO

indicate the task is ranking or rating prediction

evaluation metrics automatically selected based on True/False setting here

ranking: True

indicate activation function for ranking task

LeakyReLu is the default activation function for both ranking or rating prediction

Sigmoid : True

define metrics for ranking and rating prediction tasks

ranking_valid_metric: Recall
ranking_metrics: ['Precision','Recall','NDCG','MRR','MAP']
topk: [5,10,20]

err_valid_metric: MAE
err_metrics: ['MAE','RMSE','AUC']

############### Output setting ###############
loss_decimal_place: 4
metric_decimal_place: 4

############### Negative Sampling setting ###############
train_neg_sample_args:
strategy: 'full' # Choose a strategy (e.g., 'none', 'by', 'full')
distribution: 'uniform' # Negative sampling distribution (optional)
eval_neg_sample_args:
strategy: 'full' # Choose a strategy (e.g., 'none', 'by', 'full')
distribution: 'uniform'