streamsight.settings.SlidingWindowSetting

class streamsight.settings.SlidingWindowSetting(background_t: int, window_size: int = 2147483647, n_seq_data: int = 10, top_K: int = 10, t_upper: int = 2147483647, t_ground_truth_window: int | None = None, seed: int | None = None)

Bases: Setting

Sliding window setting for splitting data.

The data is split into a background set and evaluation set. The evaluation set is defined by a sliding window that moves over the data. The window size is defined by the window_size parameter. The evaluation set comprises of the unlabeled data and ground truth data stored in a list. The unlabeled data contains the last n_seq_data interactions of the users/item before the split point along with masked interactions after the split point. The number of interactions per user/item is limited to top_K. The ground truth data is the interactions after the split point and spans window_size seconds.

Core attribute

background_data: Data used for training the model. Interval is [0, background_t).
unlabeled_data: List of unlabeled data. Each element is a InteractionMatrix object of interval [0, t).
ground_truth_data: List of ground truth data. Each element is a InteractionMatrix object of interval [t, t + window_size).
data_timestamp_limit: List of timestamps that the splitter will slide over the data.
incremental_data: List of data that is used to incrementally update the model. Each element is a InteractionMatrix object of interval [t, t + window_size).

param background_t:: Time point to split the data into background and evaluation data. Split will be from [0, t)
type background_t:: int
param window_size:: Size of the window in seconds to slide over the data. Affects the incremental data being released to the model. If :param:`t_ground_truth_window` is not provided, ground truth data will also take this window.
type window_size:: int, optional
param n_seq_data:: Number of last sequential interactions to provide as data for model to make prediction.
type n_seq_data:: int, optional
param top_K:: Number of interaction per user that should be selected for evaluation purposes.
type top_K:: int, optional
param t_upper:: Upper bound on the timestamp of interactions. Defaults to maximal integer value (acting as infinity).
type t_upper:: int, optional
param t_ground_truth_window:: Size of the window in seconds to slide over the data for ground truth data. If not provided, defaults to window_size during computation.
type t_ground_truth_window:: int, optional
param seed:: Seed for random number generator.
type seed:: int, optional

__init__(background_t: int, window_size: int = 2147483647, n_seq_data: int = 10, top_K: int = 10, t_upper: int = 2147483647, t_ground_truth_window: int | None = None, seed: int | None = None)

Methods

`__init__`(background_t[, window_size, ...])
`destruct_generators`()	Destruct data generators.
`get_params`()	Get the parameters of the setting.
`next_ground_truth_data`([reset])	Get the next ground truth data.
`next_incremental_data`([reset])	Get the next incremental data.
`next_t_window`([reset])	Get the next data timestamp limit.
`next_unlabeled_data`([reset])	Get the next unlabeled data.
`reset_data_generators`()	Reset data generators.
`restore_generators`([n])	Restore data generators.
`split`(data)	Splits :param:`data` according to the setting.

Attributes

`background_data`	Background data provided for the model for the initial training.
`ground_truth_data`	Ground truth data to evaluate the model's predictions on.
`identifier`	Name of the setting.
`incremental_data`	Data that is used to incrementally update the model.
`is_ready`	Flag on setting if it is ready to be used for evaluation.
`is_sliding_window_setting`	Flag to indicate if the setting is `SlidingWindowSetting`.
`name`	Name of the setting.
`num_split`	Number of splits created from dataset.
`params`	Parameters of the setting.
`t_window`	The upper timestamp of the window in split.
`unlabeled_data`	Unlabeled data for the model to make predictions on.
`window_size`	Window size in seconds for splitter to slide over the data.
`t_upper`	Upper bound on the timestamp of interactions.
`n_seq_data`	Number of last sequential interactions to provide in `unlabeled_data` as data for model to make prediction.
`top_K`	Number of interaction per user that should be selected for evaluation purposes in `ground_truth_data`.

_abc_impl = <_abc._abc_data object>

_background_data: InteractionMatrix: Data used as the initial set of interactions to train the model.

_check_size(): Warns user if any of the sets is unusually small or empty

_check_split()

Checks that the splits have been done properly.

Makes sure all expected attributes are set.

_check_split_complete()

Check if the setting is ready to be used for evaluation.

Raises:: KeyError – If the setting is not ready to be used for evaluation.

_create_generator(attribute: str) → Any

Creates generator for provided attribute name

Parameters:: attribute (str) – the attribute name to be used to create the generator
Yield:: Data return from the attribute
Return type:: Any

_ground_truth_data: InteractionMatrix | List[InteractionMatrix]: Data containing the ground truth interactions to _unlabeled_data. If SlidingWindowSetting, then it will be a list of InteractionMatrix.

_ground_truth_data_generator()

Generates ground truth data.

Allow for iteration over the ground truth data. If the setting is a sliding window setting, then it will iterate over the list of ground truth data.

Note

A private method is specifically created to abstract the creation of the generator and to allow for easy resetting when needed.

_incremental_data: List[InteractionMatrix]: Data that is used to incrementally update the model. Unique to SlidingWindowSetting.

_incremental_data_generator()

Generates incremental data.

Allow for iteration over the incremental data. If the setting is a sliding window setting, then it will iterate over the list of incremental data.

Note

A private method is specifically created to abstract the creation of the generator and to allow for easy resetting when needed.

_next_t_window_generator()

Generates t_window data.

Allow for iteration over the t_window data. If the setting is a sliding window setting, then it will iterate over the list of data timestamp limit.

Note

A private method is specifically created to abstract the creation of the generator and to allow for easy resetting when needed.

_num_full_interactions: int

_split(data: InteractionMatrix)

Splits dataset into a background, unlabeled and ground truth data.

Parameters:: data (InteractionMatrix) – Interaction matrix to be split. Must contain timestamps.

_split_complete: Number of splits created from sliding window. Defaults to 1 (no splits on training set).

_t_window: None | int | List[int]: This is the upper timestamp of the window in split. The actual interaction might have a smaller timestamp value than this because this will is the t cut off value.

_unlabeled_data: InteractionMatrix | List[InteractionMatrix]

_unlabeled_data_generator()

Generates unlabeled data.

Allow for iteration over the unlabeled data. If the setting is a sliding window setting, then it will iterate over the list of unlabeled data.

Note

A private method is specifically created to abstract the creation of the generator and to allow for easy resetting when needed.

property background_data: InteractionMatrix

Background data provided for the model for the initial training.

This data is used as the initial set of interactions to train the model.

Returns:: Interaction Matrix of training interactions.
Return type:: InteractionMatrix

destruct_generators() → None

Destruct data generators.

Destructs the data generators of the setting object. This method is useful when the setting object needs to be be pickled or saved to disk.

get_params() → Dict[str, Any]: Get the parameters of the setting.

property ground_truth_data: InteractionMatrix | List[InteractionMatrix]

Ground truth data to evaluate the model’s predictions on.

Contains the actual interactions of the user-item interaction that the model is supposed to predict.

Returns:: _description_
Return type:: Union[InteractionMatrix, List[InteractionMatrix]]

property identifier: str: Name of the setting.

property incremental_data: List[InteractionMatrix]

Data that is used to incrementally update the model.

Unique to sliding window setting.

Returns:: _description_
Return type:: List[InteractionMatrix]

property is_ready: bool

Flag on setting if it is ready to be used for evaluation.

Returns:: If the setting is ready to be used for evaluation.
Return type:: bool

property is_sliding_window_setting: bool

Flag to indicate if the setting is SlidingWindowSetting.

Returns:: If the setting is SlidingWindowSetting.
Return type:: bool

n_seq_data: int: Number of last sequential interactions to provide in unlabeled_data as data for model to make prediction.

property name: str

Name of the setting.

Returns:: Name of the setting.
Return type:: str

next_ground_truth_data(reset=False) → InteractionMatrix

Get the next ground truth data.

Get the next ground truth data for the corresponding split. If the setting is a sliding window setting, then it will iterate over the list of ground truth data.

Parameters:: reset (bool, optional) – To reset the generator, defaults to False
Raises:: EOWSetting – If there is no more ground truth data to iterate over.
Returns:: The next ground truth data for the corresponding split.
Return type:: InteractionMatrix

next_incremental_data(reset=False) → InteractionMatrix

Get the next incremental data.

Get the next incremental data for the corresponding split. If the setting is a sliding window setting, then it will iterate over the list of incremental data.

Parameters:

reset (bool, optional) – To reset the generator, defaults to False

Raises:

AttributeError – If the setting is not a sliding window setting.
EOWSetting – If there is no more incremental data to iterate over.

Returns:

The next incremental data for the corresponding split.

Return type:

InteractionMatrix

next_t_window(reset=False) → int

Get the next data timestamp limit.

Get the next upper timestamp limit for the corresponding split. If the setting is a sliding window setting, then it will iterate over the list of timestamps that specify the timestamp cut off for the data.

Parameters:: reset (bool, optional) – To reset the generator, defaults to False
Raises:: EOWSetting – If there is no more data timestamp limit to iterate over.
Returns:: The next t_window for the corresponding split.
Return type:: int

next_unlabeled_data(reset=False) → InteractionMatrix

Get the next unlabeled data.

Get the next unlabeled data for the corresponding split. If the setting is a sliding window setting, then it will iterate over the list of unlabeled data.

Parameters:: reset (bool, optional) – To reset the generator, defaults to False
Raises:: EOWSetting – If there is no more unlabeled data to iterate over.
Returns:: The next unlabeled data for the corresponding split.
Return type:: InteractionMatrix

property num_split: int

Number of splits created from dataset.

This property defaults to 1 (no splits on training set) on a typical setting. Usually for the SlidingWindowSetting this property will be greater than 1 if there are multiple splits created from the sliding window on the dataset.

Returns:: Number of splits created from dataset.
Return type:: int

property params: Parameters of the setting.

reset_data_generators() → None

Reset data generators.

Resets the data generators to the beginning of the data series. API allows the programmer to reset the data generators of the setting object to the beginning of the data series.

restore_generators(n: int | None = None) → None

Restore data generators.

Restores the data generators of the setting object. If :param:`n` is provided, then it will restore the data generators to the iteration number :param:`n`. If :param:`n` is not provided, then it will restore the data generators to the beginning of the data series.

Parameters:: n (int, optional) – iteration number to restore generator to, defaults to int

split(data: InteractionMatrix) → None

Splits :param:`data` according to the setting.

Calling this method will change the state of the setting object to be ready for evaluation. The method will split the data into background_data, ground_truth_data, unlabeled_data.

This method will perform a basic check on the split to ensure that the split did not result in any empty or unusually small datasets.

Note

SlidingWindowSetting will have additional attribute incremental_data.

Parameters:: data (InteractionMatrix) – Interaction matrix that should be split.

t_upper: Upper bound on the timestamp of interactions. Defaults to maximal integer value (acting as infinity).

property t_window: None | int | List[int]

The upper timestamp of the window in split.

In settings that respect the global timeline, a timestamp value will be returned. In the case of SlidingWindowSetting, a list of timestamp values will be returned.

Settings such as LeaveNOutSetting will return None since there is no split with respect to time.

Returns:: timestamp limit for the data.
Return type:: Union[int, List[int]]

top_K: int: Number of interaction per user that should be selected for evaluation purposes in ground_truth_data.

property unlabeled_data: InteractionMatrix | List[InteractionMatrix]

Unlabeled data for the model to make predictions on.

Contains the user/item ID for prediction along with previous sequential interactions of user-item on items if it exists. This data is used to make predictions on the ground truth data.

Returns:: Either a single InteractionMatrix or a list of InteractionMatrix if the setting is a sliding window setting.
Return type:: Union[InteractionMatrix, List[InteractionMatrix]]

window_size: Window size in seconds for splitter to slide over the data.