create_train_matrices: Create training data matrix and a training response for given...

In dshemetov/modeltools-mirror: Tools for Building COVID Forecasters and Related Models

Description

Create training and data matrix and training response for a set of given aheads. Works for both single ahead values and a vector of ahead values. However, note that this function works different from create_train_and_predict_matrices() for multiple ahead values. If multiple ahead values are supplied, we return a matrix of responses containing all aheads at once.

Usage

create_train_matrices(lagged_df, ahead, n_days, train_end_date)

Arguments

lagged_df	Data frame of lagged data as in create_train_and_predict_matrices().
ahead	Number of incidence period units (i.e., epiweeks, days, etc.) ahead to forecast. Can be a single positive integer or a vector of positive integers. Note that for each {a} in ahead, the column response+{a}:{response} should be present in lagged_df.
n_days	Number of days worth of data to pull.
train_end_date	The last date to be included in the training data.

Value

A named list with entries:

• train_x: Matrix of training data whose columns correspond to the value-{days}:{signal} columns in lagged_df. The training data consists of the latest date with an non-null response, plus all data from the training_window_size days prior to it.

• train_y: Vector of response data from the response+{ahead}:{response} column of lagged_df corresponding to the rows of train_x. If multiple ahead values are provided, then this is a matrix instead.

• train_geo_values: Vector of geo values corresponding to the rows of train_x.

• train_time_values: Vector of time values corresponding to the rows of train_x.

• train_end_date: Latest time_value used in the training period.

dshemetov/modeltools-mirror documentation built on Jan. 7, 2022, 12:23 a.m.