refactor tournament wins/draws/losses, adjust displays #30
Conversation
| N = self.total_engines | ||
| len_names = max(5, min(24, max([len(x) for x in self.engine_names]) + 1)) | ||
| len_score = max(6, max([math.floor(math.log10(max(1, self.total_scores[i])) + 1) for i in range(N)]) + 1) | ||
| len_games = max(7, max([math.floor(math.log10(max(1, self.game_counts[i])) + 1) for i in range(N)]) + 1) |
There was a problem hiding this comment.
The number at the front with max(# is effectively the minimum column width. If there is then a min(# that is effectively the maximum column width, otherwise no max width. Then there is a max of the data, usually with + 1 for one additional padding on the max data length
| wins[winners[0]] += 1 | ||
| else: | ||
| for p in winners: | ||
| draws[p] += 1 |
There was a problem hiding this comment.
Distinguish between truly winning and drawing
| return self.elo_end[engine] | ||
|
|
||
| def get_delta_elo_by_engine(self, engine: int) -> int: | ||
| return self.elo_end[engine] - self.elo_start[engine] |
There was a problem hiding this comment.
These all have properties now, so just use results.property[engine] instead of results.function(engine)
| out += f"{'Avg Score':>10} {'Wins':>6} {'Win Rate':>9}\n" | ||
| ranked_engines = sorted(range(N), key=lambda x: -elos[x]) | ||
| for rank, engine in enumerate(ranked_engines): | ||
| len_name = max(5, min(24, max([len(x.name) for x in self.engines]) + 1)) |
There was a problem hiding this comment.
I'd like to not have this function duplicated, but it uses so much data it would be really annoying to pass all necessary data into a static version or something. So for now I'm most happy with it duplicated, one here in the running tournament instance and one in the results data class.
TLDR:
K=8for Elo calculations in tournaments, should be more stable/slow changingMore specifically on "winning": Now track "wins" for players truly winning alone, without a tie to another player. Then "draws" are for players tied with the highest score, and "losses" are for anyone without the highest score.
Old Summary
New Summary
Summary changes:
Results structure changes:
win_countsis now the new definition of winsdraw_countsnew property for the amount of drawslose_countsnew computed property bygame_counts - win_counts - draw_countswin_ratesunchanged computed property, but uses the new definition of winsdraw_ratesnew computed property for the rate of drawinglose_ratesnew computed property for the rate of losingwin_draw_ratesnew computed property for the old definition of winning,(win_counts + draw_counts) / game_countselo_deltanew computed property for the change in elo from the tournament