Review: SyntheticSlateBanditDataset

[aiueola]

important

1. It may be better to check the calculation of pscore_item_position_i_l:

# from
if sampled_action_index not in action_list:
# to
if sampled_action != action_list[position_]:

as pscore is calculated as :math: \\pi_k(a_k \\mid x) = \\sum_{a} \\pi(a \\mid x) \\mathbbm{1}{a(k)=a_k}.
https://github.com/aiueola/zr-obp/blob/57cae57736f00a8f4887c4d3ac41a6544b880b34/obp/dataset/synthetic_slate.py#L376

2. Also, I revised pscore_item_position_i_l when using uniform random policy (self.behavior_policy_function is None):

# from
pscore_item_position_i_l = self.len_list / self.n_unique_action
#to
pscore_item_position_i_l = 1 / self.n_unique_action

https://github.com/aiueola/zr-obp/blob/57cae57736f00a8f4887c4d3ac41a6544b880b34/obp/dataset/synthetic_slate.py#L368

refactor

1. Removed return_exact_uniform_pscore_item_position parameter from obtain_batch_bandit_feedback() and sample_action_and_obtain_pscore().

# from
if return_exact_uniform_pscore_item_position: 
# to
if self.behavior_policy_function is None:  # uniform random

It might be more intuitive that we use ground-truth pscore (not approximated one) as default for uniform random policy.
https://github.com/aiueola/zr-obp/blob/57cae57736f00a8f4887c4d3ac41a6544b880b34/obp/dataset/synthetic_slate.py#L447

2. Renamed from self.behavior_policy to self.uniform_behavior_policy as it always indicates uniform behavior policy.

# from
self.behavior_policy = np.ones(self.n_unique_action) / self.n_unique_action
# to
self.uniform_behavior_policy = np.ones(self.n_unique_action) / self.n_unique_action

https://github.com/aiueola/zr-obp/blob/57cae57736f00a8f4887c4d3ac41a6544b880b34/obp/dataset/synthetic_slate.py#L245

3. Renamed from action_interaction_matrix to action_interaction_weight_matrix.

https://github.com/aiueola/zr-obp/blob/35f893fd660f627b0bb470d24eaf970aeb541109/obp/dataset/synthetic_slate.py#L49

tests

Fix pscore_item_position target of uniform random policy:

# from 
pscore_item_position = len_list / n_unique_action
#to
pscore_item_position = 1 / n_unique_action

https://github.com/aiueola/zr-obp/blob/d46cd4c9a626020a14c73014a21b8749b9dd88ad/tests/dataset/test_synthetic_slate.py#L229

others

Minor fix on typos and docstrings.