Feature: Implementing SyntheticContinuousBanditDataset

[usaito]

new feature

• implement SyntheticContinuousBanditDataset that generates synthetic data with (1-dimensional) continuous actions
  https://github.com/st-tech/zr-obp/blob/continuous-dataset/obp/dataset/synthetic_continuous.py

• this class works as follows

from obp.dataset import (
    SyntheticContinuousBanditDataset,
    linear_reward_funcion_continuous,
    linear_behavior_policy_continuous,
)

dataset = SyntheticContinuousBanditDataset(
    dim_context=5,
    min_action_value=1, 
    max_action_value=10,
    reward_function=linear_reward_funcion_continuous,
    behavior_policy_function=linear_behavior_policy_continuous,
    random_state=12345,
)
bandit_feedback = dataset.obtain_batch_bandit_feedback(n_rounds=10000)

• Note that by setting min_action_value and max_action_value, we can control the action space (\mathcal{A} = [1,10] in the above example code)

tests

• add some tests of SyntheticContinuousBanditDataset
  https://github.com/st-tech/zr-obp/blob/continuous-dataset/tests/dataset/test_synthetic_continuous.py

refactor

• rename action_prob to pscore in test_synthetic.py
  

  zr-obp/tests/dataset/test_synthetic.py

  Lines 313 to 315 in c77dd99

  	pscore = linear_behavior_policy(context=context, action_context=action_context)
  	assert pscore.shape[0] == n_rounds and pscore.shape[1] == n_actions
  	assert np.all(0 <= pscore) and np.all(pscore <= 1)

[nomuramasahir0]

Other than the above minor points, LGTM!

[usaito]

@nmasahiro Thanks!