BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//GYYTCH
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-euroscipy-2023-GYYTCH@pretalx.com
DTSTART;TZID=CET:20230816T110500
DTEND;TZID=CET:20230816T113500
DESCRIPTION:When operating a classifier in a production setting (i.e. predi
ctive phase)\, practitioners are interested in potentially two different o
utputs: a "hard" decision used to leverage a business decision or/and a "s
oft" decision to get a confidence score linked to each potential decision
(e.g. usually related to class probabilities).\n\nScikit-learn does not pr
ovide any flexibility to go from "soft" to "hard" predictions: it uses a c
ut-off point at a confidence score of 0.5 (or 0 when using `decision_funct
ion`) to get class labels. However\, optimizing a classifier to get a conf
idence score close to the true probabilities (i.e. a calibrated classifier
) does not guarantee to obtain accurate "hard" predictions using this heur
istic. Reversely\, training a classifier for an optimum "hard" prediction
accuracy (with the cut-off constraint at 0.5) does not guarantee obtaining
a calibrated classifier.\n\nIn this talk\, we will present a new scikit-l
earn meta-estimator allowing us to get the best of the two worlds: a calib
rated classifier providing optimum "hard" predictions. This meta-estimator
will land in a future version of scikit-learn: https://github.com/scikit-
learn/scikit-learn/pull/26120.\n\nWe will provide some insights regarding
the way to obtain accurate probabilities and predictions and also illustra
te how to use in practice this model on different use cases: cost-sensitiv
e problems and imbalanced classification problems.
DTSTAMP:20241010T201719Z
LOCATION:HS 120
SUMMARY:Get the best from your scikit-learn classifier: trusted probabiltie
s and optimal binary decision - Guillaume Lemaitre
URL:https://pretalx.com/euroscipy-2023/talk/GYYTCH/
END:VEVENT
END:VCALENDAR