Zindi
CompeteJobsLearnChatLeaderboardFor Organisations
PartnersMeet the teamPressCase studiesAI4EAC
Primary competition visual

Basic Needs Basic Rights Kenya - Tech4MentalHealth

Helping Kenya
$4 200 USD
Completed (over 6 years ago)
Classification
1257 joined
499 active
InfoDataChatLeaderboard
Starti
26 Apr 20
Closei
05 Jul 20
Reveali
05 Jul 20
User avatar
c
evaluation metrics and task clarification
Help · 3 Jul 2020, 21:45 · 7

It is stated that evaluation metrics is log-loss and at the same it is mentioned that probability values do not have to add up to one. That's why it seems like the task is multi-label classification, so it is possible that true target on the test dataset might have more than one label or it is not?

Discussion 7 answers
User avatar
Satoshi
3 Jul 2020, 22:20

can anyone please help, am having issues calculating log loss.

The error states that; y_true and y_pred contain different number of classes 4, 2. Please provide the true labels explicitly through the labels argument. Classes found in y_true: [0 1 2 3].

I have substituted the classes with the labels, but to no avail, the code didn't run. Can you please help.

Upvotes 0
User avatar
devnikhilmishra
4 Jul 2020, 02:13replied to Satoshi

Hi Zion, use labels argument in log_loss(), and do log_loss(y_true, y_preds, labels=[0,1,2,3]). Then it will work.

Upvotes 0
User avatar
Satoshi
4 Jul 2020, 10:39replied to devnikhilmishra

Alright. Thanks

Upvotes 0
User avatar
asingleneuron
4 Jul 2020, 03:34

@c , I treated this as a classification because in training dataset non of the data point belongs to more than one class. So if you don't have multi-labeled training data, we can't train a multi-label ML model.

Upvotes 0
User avatar
Krishna_Priya
4 Jul 2020, 16:26replied to asingleneuron

Hey, this can be treated as a multi-label problem as there are few same and many similar statements having different labels, but I tried it initially and the cv and LB scores were worse. So I left this idea but looking at the data set it was my first intuition. Hoping to hear from someone who was successful in this approach after the competition.

Upvotes 0
User avatar
asingleneuron
4 Jul 2020, 17:21replied to Krishna_Priya

But the success is relying on if the test dataset labeled as multi-label if not then chances of success are less in my opinion.

Upvotes 0
User avatar
Krishna_Priya
5 Jul 2020, 06:00replied to asingleneuron

Yes, you are right.

Upvotes 0

Join the largest network for
data scientists and AI builders

AboutFAQsPrivacy PolicyTerms of UseRules
Status
© Zindi 2026