You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
negate the result and prefix the metric name for error/loss metrics (#278)
* negate the result and prefix the metric name for error/loss metrics
* use the occasion to document available metrics and add more of them
* docstring
* fixed reports library+notebook to accommodate the changes
* small cleanup in reports notebook
* fixed wrong search/replace in reporting notebook
Copy file name to clipboardExpand all lines: amlb_report/results.py
+24-41Lines changed: 24 additions & 41 deletions
Original file line number
Diff line number
Diff line change
@@ -1,11 +1,12 @@
1
1
"""
2
-
Loading results, formatting and adding columns
3
-
result is the raw result metric computed from predictions at the end the benchmark. For classification problems, it is usually auc for binomial classification and logloss for multinomial classification.
4
-
score ensures a standard comparison between tasks: higher is always better.
5
-
norm_score is a normalization of score on a [0, 1] scale, with {{zero_one_refs[0]}} score as 0 and {{zero_one_refs[1]}} score as 1.
6
-
imp_result and imp_score for imputed results/scores. Given a task and a framework:
7
-
if all folds results/scores are missing, then no imputation occurs, and the result is nan for each fold.
8
-
if only some folds results/scores are missing, then the missing result is imputed by the {{imp_framework}} result for this fold.
2
+
Loading results, formatting and adding columns.
3
+
result is the raw result metric computed from predictions at the end the benchmark: higher is always better!
4
+
- For classification problems, it is usually auc for binary problems and negative log loss for multiclass problems.
5
+
- For regression problems, it is usually negative rmse.
6
+
norm_result is a normalization of result on a [0, 1] scale, with {{zero_one_refs[0]}} scoring as 0 and {{zero_one_refs[1]}} scoring as 1.
7
+
imp_result for imputed results. Given a task and a framework:
8
+
- if all folds results are missing, then no imputation occurs, and the result is nan for each fold.
9
+
- if only some folds results are missing, then the missing result is imputed by the {{imp_framework}} result for this fold.
0 commit comments