Evaluation Parameters

Required Evaluation Parameters

Name	Type	Description
leftKeys	list[str]	List of column names for the left features, typically text fields.
leftWeights	list[str]	List of weights for the left features.
rightKeys	list[str]	List of column names for the right features. The first field should be an image, and the others should be text fields.
rightWeights	list[float]	List of weights for the right features.

Optional Evaluation Parameters

Name	Type	Default value	Description
batchSize	int	64	Batch size.
contextLength	int	77	Maximum number of tokens in the input text to train with.
numWorkers	int	4	Number of data loader workers.
runQueriesCpu	bool	True	Run queries in CPU (faster and less prone to GPU memory overflows).
topQ	int	4000	The number of queries to sample from the input set. Must be between 1 and 20,000. If the specified value exceeds the number of unique queries in the dataset, it will be automatically adjusted to match the total unique queries available.
weightKey	str	marqtune__score	The column name in the dataset that contains the score. Defaults to marqtune__score which is constant score value 1 for all rows.

Example

evaluate_task_params = {
    "leftKeys": ["query"],
    "rightKeys": ["my_image", "my_text"],
    "leftWeights": [1],
    "rightWeights": [0.9, 0.1],
    "batchSize": 32,
    "numWorkers": 4,
    "contextLength": 77,
    "topQ": 2000,
}