Run Predictions
The overview of running predictions
Once you train downstream model, you probably wish to make predictions. In order to do that, you must:
- Load trained downstream model using
load_from_checkpointmethod. We load the best model according to the specified metric, defined during training. - Define testing parameters in
MonadTestingParamsclass. You can overwrite parameters configured for training of downstream model.
checkpoint_dir = "<path/to/downstream/model/checkpoints>"
testing_module = load_from_checkpoint(checkpoint_dir)
Creates MonadModuleImpl from MonadCheckpoint.
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| checkpoint_path | str | Directory where all the checkpoint artifacts are stored. | required |
| pl_logger | Optional[Logger] | An instance of PyTorch Lightning logger to use. | None |
| loading_config | Optional[LoadingConfigParams] | A dictionary containing a mapping from datasource name (or from datasource name and mode) to the fields of DataSourceLoadingConfig. If provided, the listed parameters will be overwritten. Field datasource_cfg can't be changed. | None |
| kwargs | Data parameters to change. | {} |
Returns
| Name | Type | Description |
|---|---|---|
| MonadModuleImpl | MonadModuleImpl | Instance of monad module, loaded from the checkpoint. |
Additonally, you can pass any parameters defined in MonadDataParams in order to overwrite parameters configured for training of downstream model:
Good to know
It is in this module, that you will define the prediction window, i.e. the time window that you want to predict for your target function.
For example, if you plan to predict the propensity to purchase something within 21 days from a given date, you need to define this using test_start_date and check_target_for_next_N_days = 21
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| features_path | str | A path to the folder with features created during the pretrain phase. | required |
| data_start_date | datetime | Events after this date will be considered for training. | required |
| check_target_for_next_N_days | int | The number of days used to create the model's target. Not suitable for recommendation models. | None |
| validation_start_date | datetime | Start date for the validation set. | None |
| test_start_date | datetime | The date that the prediction is being calculated for. validation_start_date or test_start_date needs to be provided. | None |
| test_end_date | datetime | End date of the test period. | None |
| timebased_encoding | str | How to encode time-based features; available encoding options are "fourier" or "two-hot". | 'two-hot' |
| target_sampling_strategy | str | "valid" or "random" sampling strategy. For Foundation Model, it should always be "random". | 'random' |
| maximum_splitpoints_per_entity | int | The maximum number of splits into input and target events per entity. | 1 |
| num_query_chunks | int | The number of segments a query should be divided into to reduce memory consumption on the database end. | 1 |
| use_recency_sketches | boolean | If true, then recency sketches are used in training. | True |
Then, instantiate monad.core.config.MonadTestingParams with the provided parameters. If not specified, they will be overwritten with parameters from the downstream training module.
from monad.ui.config import MonadTestingParams
testing_params = MonadTestingParams(
save_path="<path/to/save/predictions>"
)
Parameters
| Name | Type | Description | Default |
|---|---|---|---|
| save_path | str | If provided, points to the location where predictions will be stored in CSV format. | required |
| limit_test_batches | Optional[Union[int, float]] | How much of the test dataset to check (float = fraction, int = num_batches). | None |
| devices | Union[List[int], str, int, None] | The devices to use. Can be set to a positive number (int or str), a sequence of device indices(list or str), the value -1 to indicate all available devices should be used, or auto for automatic selection based on the chosen accelerator. | field(default_factory=lambda : [0]) |
| accelerator | str | The accelerator to use, as in PyTorch Lightning trainer. | 'gpu' |
| precision | Literal[64, 32, 16, '64', '32', '16', 'bf16', '16-true', '16-mixed', 'bf16-true', 'bf16-mixed', '32-true', '64-true'] | Double precision, full precision, 16bit mixed precision or bfloat16 mixed precision. | DEFAULT_PRECISION |
| metrics | Dict[str, Metric] | Metrics to use in validation. If not provided, default validation metrics function for a task will be used. | None |
| callbacks | List[Callback] | List of additional callbacks to add to validation/testing. | field(default_factory=list) |
| top_k | int | Only for recommendation task. Number of targets to recommend. Top k targets will be included in predictions. | 12 |
| targets_to_include | List[str] | Only for recommendation task. Target names that will be included in predictions. | None |
and finally make predictions:
testing_module.predict(testing_params=testing_params)
Did you know?
You can use subqueries to filter users that you run predictions on. For example to only calculate propensity for product A for users that never purchase this particular product and get the list of users with highest propensity to purchase it.
To make use of this functionality you need to use set_entities_ids_subqueryaccessible from load_from_checkpoint and load_from_foundation_model methods and provide a SQL query in the flavor corresponding to the database you are using.
Updated 2 months ago