Kaggle's 30 Days Of ML (Competition Part-4): Hyperparameter tuning using Optuna

No video

Kaggle's 30 Days Of ML (Competition Part-4): Hyperparameter tuning using Optuna

Рет қаралды 11,715

Abhishek Thakur

Күн бұрын

Пікірлер: 34

@abhishekkrthakur 3 жыл бұрын

Notebook-1: www.kaggle.com/abhishek/competition-part-4-hyperparameter-tuning-optuna Notebook-2: www.kaggle.com/abhishek/competition-part-4-optimized-xgboost Please subscribe and like the video to help me keep motivated to make awesome videos like this one. :)

@channelname9332 3 жыл бұрын

🔥👍

@geekyprogrammer4831 3 жыл бұрын

Thank God you are back 😭😭😭 Once I get a job related to DS, definitely will pay you back brother!

@madhu1987ful 3 жыл бұрын

Woaaah !!! after this optuna tuning I jumped from 1500 rank to top 300 !!! :-)

@abhishekkrthakur 3 жыл бұрын

Nice work! Watch next one and jump to 30!

@vivekchowdhury4225 2 жыл бұрын

Beautifully explained GradMaster🙌 I was searching for a tutorial like this for a long time. Also, your book is amazing.

@soumyasubhrabhowmik2209 3 жыл бұрын

Thank you so much for your clear explanations of what I believe are pretty complex concepts. It has been a great experience learning from you over the past few weeks.

@agamenon1953 3 жыл бұрын

Abhishek, you're amazing! Thank you so much for sharing this valuable knowledge!

@t-m5678 3 жыл бұрын

Thanks for the videos. These are very helpful.

@floopybits8037 2 жыл бұрын

Nice explanation of optuna , i giggled when you said n_estimators =7000 is small since i have a 8GB PC 😁

@loguansiang2300 3 жыл бұрын

Thanks for the videos.Hope u can do more series for other tutorial and competition.

@pandasaspd6588 2 жыл бұрын

Thank you very much, I learned a lot from this video

@Yu-nd1kr Жыл бұрын

hi if i use cnn1d model what code if i use optuna for cnn1d to optimize filter and kernal ??

@priyanksharma1200 3 жыл бұрын

What does log=true do in suggest_float? How is it different from suggest_loguniform or suggest_discrete_uniform?

@eyalbaum1254 3 жыл бұрын

love the video! was wondering - why not incorporating cross_val_score within the study? won't it deliver better results in terms of model selction? i tried my best to incorporate it but couldn't find an elegent solution (maybe i don't even have to)

@abhishekkrthakur 3 жыл бұрын

because we created our own folds in part-1 and we have been using them. cross_val_score creates its own folds but we want to use our own folds :)

@kiranchowdary8100 3 жыл бұрын

what is the reason for models giving high accuracy with cpu and not so high or a bit less with gpU? any technical reasons to know?

@abhishekkrthakur 3 жыл бұрын

check out default value of tree_method parameter for gpu and cpu.

@AI-Kawser 3 жыл бұрын

Sir can you please share the link to the other video you mentioned here in the beginning. Thank you.

@abhishekkrthakur 3 жыл бұрын

kzfaq.info/get/bejne/a9SJpK5ercfTe40.html&ab_channel=AbhishekThakur

@madhu1987ful 3 жыл бұрын

Hi Abhishek, In optuna code, i observed that u have commented the GPU params in XGBRegressor while submitting predictions, but same were enabled during hyperparam tuning. Is there a reason?

@abhishekkrthakur 3 жыл бұрын

the answer to the question is already in my video :) gpu uses `tree_method="gpu_hist"` but cpu uses `tree_method="exact"`. apparently, the latter is better on this dataset. so we finetune on gpu, so that it is fast and submit with cpu.