Predict API

Once your model is trained, you must deploy the model on Abacus.AI platform to generate predictions. You can use the prediction dashboard to generate the predictions from the trained model. In this section the underlying prediction API and all other additional prediction API methods are discussed for the use case in consideration:

Method
getRankedItems POST
Copy POST

Returns a list of re-ranked items for a selected user when a list of items is required to be reranked according to the user's preferences. Note that the inputs to this method, wherever applicable, will be the column names in your dataset mapped to the column mappings in our system (e.g. column 'item_code' mapped to mapping 'ITEM_ID' in our system).

Arguments:

REQUIRED KEY TYPE DESCRIPTION
Yes deploymentToken str The deployment token to authenticate access to created deployments. This token is only authorized to predict on deployments in this project, so it is safe to embed this model inside of an application or website.
Yes deploymentId str The unique identifier to a deployment created under the project.
Yes queryData dict This will be a dictionary with two key-value pairs. The first pair represents a 'Key' where the column name (e.g. a column with name 'user_id' in your dataset) mapped to the column mapping USER_ID uniquely identifies the user against whom a prediction is made and a 'Value' which is the identifier value for that user. The second pair will have a 'Key' which will be the name of the column name (e.g. movie_name) mapped to ITEM_ID (unique item identifier) and a 'Value' which will be a list of identifiers that uniquely identifies those items.
No preserveRanks list List of dictionaries of format {"column": "col0", "values": ["value0, value1"]}, where the ranks of items in query_data is preserved for all the items in "col0" with values, "value0" and "value1". This option is useful when the desired items are being recommended in the desired order and the ranks for those items need to be kept unchanged during recommendation generation.
No preserveUnknownItems bool If true, any items that are unknown to the model, will not be reranked, and the original position in the query will be preserved
No scoreField str The relative item scores are returned in a separate field named with the same name as the key (score_field) for this argument.
No scalingFactors list It allows you to bias the model towards certain items. The input to this argument is a list of dictionaries where the format of each dictionary is as follows: {"column": "col0", "values": ["value0", "value1"], "factor": 1.1}. The key, "column" takes the name of the column, "col0"; the key, "values" takes the list of items, "["value0", "value1"]" in reference to which the model recommendations need to be biased; and the key, "factor" takes the factor by which the item scores are adjusted. Let's take an example where the input to scaling_factors is [{"column": "VehicleType", "values": ["SUV", "Sedan"], "factor": 1.4}]. After we apply the model to get item probabilities, for every SUV and Sedan in the list, we will multiply the respective probability by 1.1 before sorting. This is particularly useful if there is a type of item that might be less popular but you want to promote it or there is an item that always comes up and you want to demote it.
No diversityAttributeName str item attribute column name which is used to ensure diversity of prediction results.
No diversityMaxResultsPerValue int maximum number of results per value of diversity_attribute_name.
Note: The arguments for the API methods follow camelCase but for Python SDK underscore_case is followed.

Response:

KEY TYPE DESCRIPTION
success Boolean true if the call succeeded, false if there was an error
RankedItemsPrediction

Exceptions:

TYPE WHEN
DataNotFoundError

deploymentId is not found.
Language:

© Copyright