Again, much appreciated, I know I would've been lost without your videos and the same for countless others.

Here's the thing man, when i thought of quitting it ur videos came to rescue.. Learning from a grandmaster helped a lot.. Really appreciate your efforts.. Keep posting more informative videos like this... P.s - also if u have any guidance related thing for a beginner or intermediary by you, whether it could be a video or article or blog or a comment for me please give me... It would really help me to be you...

20:02 Thank you for your answer to my comment on the day 10 tutorial. I also did a quick test which proves that in a small dataset the best params is not still the best param for whole data: - your first submission when you try rf with 700 estimators - I tried from [100,1000,100] and find the best param is 900(hope that I remember correctly) based on the validation set. Then train the with 900 and 500 with the whole data than 900 get the lower score on Kaggle submission 😃 Hope that help other learners I have another question: - method 1: keep the k models trained on k fold CV. Submission: Average result of k models - method2: keep the best-param when evaluating on k folds then retrain with all data, playing around with different seeds. Then average the result. Which method do you prefer to do first in competition?

thank you abhishekh

abhishek can we take someone else model and do a tunning to it? as model building is a tough job to do

Hi, In CV we use StratifiedKFold in order to keep same ratio of target labels in each fold/split. So that when we measure our model performance, it correctly represents model stability rather getting affected by unbalanced splits. Am I correct?

15:57 Here why did you wrote n_estimators = n_estimators. I mean wouldn't n_estimators , alone be suffice