Removing constant & Quasi constant features using Variance Threshold | Machine Learning

Поділитися
Вставка
  • Опубліковано 20 гру 2024

КОМЕНТАРІ • 22

  • @raviyadav2552
    @raviyadav2552 3 роки тому +3

    The quality of content is so good, I wonder why there are fewer views. But no worries you are doing a great job bro, keep it up and people will eventually find you.

  • @leemeiwah
    @leemeiwah 3 місяці тому

    Thank you for the very clear explanation 🙂

  • @tusgyu851
    @tusgyu851 3 роки тому +1

    Hi Rachit ..while training part my model perform well.when there is class distribution like 1:60% and 0:40%.
    But when i use test data set with different class distribution like 1:30% and 0:70% then it perform worse..why? ..

    • @tusgyu851
      @tusgyu851 3 роки тому

      can you please help me with this?

    • @rachittoshniwal
      @rachittoshniwal  3 роки тому +1

      Different test sets would give you different results. So it's better to do cross validation, which provides a more holistic view of the data at hand

    • @tusgyu851
      @tusgyu851 3 роки тому

      @@rachittoshniwal Thanks for reply Rachit🙏, Yes but I have already used stratified K fold cross validation technique tried with fold 3,4. Bt still performance is bad when i change distribution for test data.

    • @tusgyu851
      @tusgyu851 3 роки тому

      Also my data set was right skewed and I also used log transformation for that.

    • @rachittoshniwal
      @rachittoshniwal  3 роки тому +1

      @@tusgyu851 cross validation will give you a general sense of the data performance. Train test split will sometimes give you good performance, sometimes bad depending on what data ends up in the test set, which makes it unreliable

  • @xristoskou6357
    @xristoskou6357 2 роки тому

    Rachit, the content of you channel is pure GOLD, thank you for sharing your knowledge, you are great!!
    However, I've got a small objection regarding the Variance threshold. Shouldn't we also check the mean of the column before we proceed with dropping a quasi-constant feature?
    For example, if we have variance = 0.0095, and a mean value of 0.003, can we still drop that feature? (lets say we have medical data)
    And question number 2 :)
    Which courses/books/youtubeChannes would you recommend for more intermediate level content like yours?
    Once again, a HUGE thank you

  • @whyme6543
    @whyme6543 3 роки тому +2

    Great video bro, can you add subtitle English?

    • @rachittoshniwal
      @rachittoshniwal  3 роки тому +1

      Sorry bro, won't be able to do it

    • @whyme6543
      @whyme6543 3 роки тому +1

      @@rachittoshniwal ok bro, but can you add automatic english subtitles on the next video

    • @rachittoshniwal
      @rachittoshniwal  3 роки тому +1

      @@whyme6543 oh, I thought the automatic ones did appear. I'll look into it though, thanks!

    • @whyme6543
      @whyme6543 3 роки тому +1

      @@rachittoshniwal thanks for your attention bro, but the automatic subtitles don't exist yet, thanks again bro

  • @zk321
    @zk321 Рік тому +1

    The word “namaste" in Sanskrit means “bowing to you". Muslims believe that one can bow/prostrate only to Allah. We don't bow down to any human. It's important to note that religious beliefs and practices can vary among individuals and communities.

  • @tableaujedi
    @tableaujedi 3 роки тому

    ..