Page MenuHomePhabricator

Train/test `reverted` model for nowiki
Closed, ResolvedPublic

Event Timeline

Extracting features for 40K revisions right now in ores-compute-01

2016-06-04 23:27:38,260 INFO:revscoring.utilities.train_test -- Training model...
2016-06-04 23:28:27,327 INFO:revscoring.utilities.train_test -- Testing model...
ScikitLearnClassifier
 - type: RF
 - params: warm_start=false, n_estimators=640, max_leaf_nodes=null, center=true, max_features="log2", scale=true, balanced_sample_weight=true, min_weight_fraction_leaf=0.0, random_state=null, n_jobs=1, verbose=0, class_weight=null, balanced_sample=false, min_samples_leaf=3, bootstrap=true, max_depth=null, criterion="entropy", oob_score=false, min_samples_split=2
 - version: 0.1.0
 - trained: 2016-06-04T23:28:27.320660

Table:
                 ~False    ~True
        -----  --------  -------
        False      7830       42
        True         51       68

Accuracy: 0.988
Precision: 0.618
Recall: 0.571
PR-AUC: 0.621
ROC-AUC: 0.981
Recall @ 0.1 false-positive rate: threshold=0.837, recall=0.134, fpr=0.059
Filter rate @ 0.9 recall: threshold=0.168, filter_rate=0.962, recall=0.908
Filter rate @ 0.75 recall: threshold=0.366, filter_rate=0.979, recall=0.756