blp_optimization_example.py

import numpy as np

from active_learning.blp_selection_utils import mip_single_task_selection, mip_unrestricted_disjoint_sets, \
    mip_equal_budget_disjoint_sets, mip_joint_task_selection, mip_single_task_confidence_selection

def main():
    # Set task names and task length
    tasks = ['deps', 'ner']
    num_tasks = len(tasks)

    # Set the total cost for our budget
    total_cost = 25

    # Randomly generate confidence scores for all tasks ranging from 0 to 1.
    rand_confidences = np.random.uniform(0, 1, (1000, num_tasks))
    # Build the mapping from samples to their confidence scores
    sample_to_confidences = dict(enumerate(rand_confidences))

    # Randomly generate costs for all tasks ranging from 0 to 1.
    rand_costs = np.random.uniform(0, 1, (1000, num_tasks))
    # Add the joint costs (in this case the average of all tasks) to the final column
    rand_costs = np.hstack((rand_costs, np.expand_dims(np.mean(rand_costs, axis=1), axis=1)))
    # Build the mapping from samples to their confidence scores
    sample_to_costs = dict(enumerate(rand_costs))

    # Get sample ids for single-task selection
    st_sample_ids = mip_single_task_selection(sample_to_confidences,
                                              sample_to_costs,
                                              total_cost,
                                              [tasks[0]])

    # Get sample ids for unrestricted disjoint sets
    udjs_sample_ids = mip_unrestricted_disjoint_sets(sample_to_confidences,
                                                     sample_to_costs,
                                                     total_cost,
                                                     tasks)

    # Get sample ids for equal budget disjoint sets
    eqbdjs_sample_ids = mip_equal_budget_disjoint_sets(sample_to_confidences,
                                                       sample_to_costs,
                                                       total_cost,
                                                       tasks)

    # Get sample ids for joint-task selection
    jt_sample_ids = mip_joint_task_selection(sample_to_confidences,
                                             sample_to_costs,
                                             total_cost,
                                             tasks)

    # Get sample ids for single-task confidence selection
    stc_sample_ids = mip_single_task_confidence_selection(sample_to_confidences,
                                                          sample_to_costs,
                                                          total_cost,
                                                          tasks,
                                                          tasks[0])

if __name__ == "__main__":
    main()