Generators of the ShareShift test points. #111
Conversation
| This class generates p * m testing points v with: | ||
| v[i] = s_i + s_i * a_k and | ||
| v[j] = s_j - s_j / (sum(s) - s[i]) * (s_i * a_k) for j != i, | ||
| iterating over 1 <= i <= p and k = 1, ..., m. Here, a_1, ..., a_m are a list |
| dataset: The DataSet for which test points are to be generated. | ||
| campaign_spend_fractions: The campaign_spend / inventory_spend | ||
| at each publisher, where each inventory_spend is given by | ||
| dataset._max_spends. |
There was a problem hiding this comment.
Similar to what Rieman recently suggested, probably add the dimension of each input argument?
There was a problem hiding this comment.
Good point, added. And also revised the description of this arg.
| v = self._campaign_spends.copy() | ||
| shift_amount = v[selected_pub] * shift_fraction | ||
| if sum(v) - v[selected_pub] > 0: | ||
| shift_proportions = v / (sum(v) - v[selected_pub]) |
There was a problem hiding this comment.
Does this equation for um(v) - v[selected_pub] > 0 first appear? Based on the description in Returns, there seems no this definition?
There was a problem hiding this comment.
Thanks, I've now called it our in the description.
| [1 / (self._npublishers - 1)] * self._npublishers | ||
| ) | ||
| v += shift_proportions * shift_amount | ||
| v[selected_pub] = self._campaign_spends[selected_pub] - shift_amount |
There was a problem hiding this comment.
After carefully reading, I can tell the codes are correct. The only issue is the codes might not be very easy to read.
There was a problem hiding this comment.
Agreed. Hope it's better now with the updated descriptions.
| for shift_fraction in self._shift_fraction_choices: | ||
| point = self.one_testing_point(selected_pub, shift_fraction) | ||
| if not any(point < 0): | ||
| yield point |
There was a problem hiding this comment.
Based on the definition in the previous function, the entries in point can be positive or negative. But here, you only care about the cases where all the entries of point are non-negative. Am I right?
There was a problem hiding this comment.
Exactly. We discard test points with any negative spend.
| Suppose that the original spend vector is [s_1, ..., s_p]. | ||
| This class generates p * m testing points v with: | ||
| v[i] = s_i + s_i * a_k and | ||
| v[j] = s_j - s_j / (sum(s) - s[i]) * (s_i * a_k) for j != i, |
There was a problem hiding this comment.
The notation here seems a bit confused. In particular, v is still a p-dimensional vector. Given an a_k, we obtain p entries, i.e., v and s have the same dimension. Then enumerating over a_1, a_2, ..., a_m, we obtain a total of pm data points. It is better not to say pm testing points v, because usually it means v has p*m entries.
There was a problem hiding this comment.
Thanks Chenwei. Motivated by your comments, I've updated the description of this class. Please take another look.
| Suppose that the original spend vector is [s_1, ..., s_p]. | ||
| This class generates p * m testing points v with: | ||
| v[i] = s_i + s_i * a_k and | ||
| v[j] = s_j - s_j / (sum(s) - s[i]) * (s_i * a_k) for j != i, |
There was a problem hiding this comment.
Thanks Chenwei. Motivated by your comments, I've updated the description of this class. Please take another look.
| dataset: The DataSet for which test points are to be generated. | ||
| campaign_spend_fractions: The campaign_spend / inventory_spend | ||
| at each publisher, where each inventory_spend is given by | ||
| dataset._max_spends. |
There was a problem hiding this comment.
Good point, added. And also revised the description of this arg.
| v = self._campaign_spends.copy() | ||
| shift_amount = v[selected_pub] * shift_fraction | ||
| if sum(v) - v[selected_pub] > 0: | ||
| shift_proportions = v / (sum(v) - v[selected_pub]) |
There was a problem hiding this comment.
Thanks, I've now called it our in the description.
| [1 / (self._npublishers - 1)] * self._npublishers | ||
| ) | ||
| v += shift_proportions * shift_amount | ||
| v[selected_pub] = self._campaign_spends[selected_pub] - shift_amount |
There was a problem hiding this comment.
Agreed. Hope it's better now with the updated descriptions.
| for shift_fraction in self._shift_fraction_choices: | ||
| point = self.one_testing_point(selected_pub, shift_fraction) | ||
| if not any(point < 0): | ||
| yield point |
There was a problem hiding this comment.
Exactly. We discard test points with any negative spend.
No description provided.