Writer Configuration¶

The Writer Configuration module defines how and where WarpRec saves the output of experiments. This includes generated recommendations, evaluation metrics, and data splits created during preprocessing.

By customizing file formats, paths, and naming conventions, WarpRec allows precise control of output organization. This feature is especially useful when running multiple experiments simultaneously or when managing several datasets.

Available Keywords¶

The following keywords are available for configuring the writer:

dataset_name: The name of the dataset currently being used. This value will be used in naming output directories and files to ensure traceability.
writing_method: The strategy used to write output files:
- local -- Writes data to local filesystem.
- azure_blob -- Writes data to an Azure Blob Storage. Needs remote-io extra to be installed.
local_experiment_path: The directory path where experiment files will be stored. This parameter is required when using the local writing method.
azure_blob_experiment_container: The name of the Azure Blob container where experiment files will be stored. This parameter is required when using the azure_blob writing method.
save_split: Whether to save the train/validation/test splits generated during preprocessing. Defaults to False.
results: A nested section controlling the formatting of result files.
split: A nested section controlling the formatting of split files.
recommendation: A nested section controlling the formatting of recommendation files.

Important

Consistency between Reader and Writer configurations is crucial. For example, if custom labels are defined in the reader, they must be mirrored in the writer to ensure correct file generation.

Results¶

The results nested section defines how result files (e.g., evaluation metrics) are formatted:

sep: Column separator for output files. Default: \t.
ext: Extension for output files. Default: .tsv.

Split¶

The split nested section configures how data split files are saved:

sep: Column separator for split files. Default: \t.
ext: Extension for split files. Default: .tsv.
header: Boolean flag indicating whether the first row should contain column names. Defaults to True.
file_format: The format of the split files. Supported values are tabular and parquet.

Recommendation¶

The recommendation nested section defines the structure of recommendation output files:

sep: Column separator for output files. Default: \t.
ext: Extension for output files. Default: .tsv.
header: Whether to include column headers in the output. Defaults to True.
k: Number of recommendations generated per user. Defaults to 50.
user_label: Label used for the user column. Defaults to user_id.
item_label: Label used for the item column. Defaults to item_id.
rating_label: Label used for the recommendation score. Defaults to rating.

Note

The k parameter controls the top-k recommendations stored per user. Increasing this value may lead to larger files and slower post-processing.

Example Configuration¶

The following example shows a complete writer configuration that saves experiment results locally with customized formatting:

writer:
    dataset_name: movielens
    writing_method: local
    local_experiment_path: results/movielens_experiment
    save_split: true
    results:
        sep: ','
    split:
        sep: ','
        ext: .csv
    recommendation:
        sep: ','
        ext: .csv
        header: false
        k: 100