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'