Thanks for the video, it's very well organized. However, you pinned down the best model, without doing some cross validation and hyperparameter tuning, that is crucial for ML models to avoid overfitting. Therefore picking the best model in this way can be misleading. You have chosen XGBoost, that might overfit the data. IMO would be really informative and useful if you could do a second part in which you show also how to tune XGboost with gridserch, optuna and bay. opt. Furthermore, a video with a comparison with XGBOOST; CATBOOST AND LIGHT GBM with their hyp tun will be very appreciated! Really looking forward to see the second part more "advanced" of this video. Thanks in advance.
@simonebenzi4189 Thanks alot for your comment. All models are validated with cross validation. Perhaps you missed it. There is already Hyperparameter Tunning with Hyperopt for XGBOOST and GridSearch of ADABOOST and Logistics Regression in my other video. You can go that specific chapter. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-i1L7qAV-_rY.htmlsi=0qnLLCOE5DYYI7_f. Hyperparameter Tuning on all models would make the video insanely long but as you suggest I make a second half. And comparing XGBOOST CATBOOST and Light GBM with hyper parameters is now I am on list after your request 👋 Let me know if you have other questions
Welcome to RU-vid! 🎉 I'm thrilled to have you here, especially as my best follower on LinkedIn. Your insightful projects and sharing of knowledge has truly impressed me. Thanks a lot for the fantastic session! I've subscribed and can't wait for more awesome content. Thank you so much!
Excellent informative video, i request you to make a end-to-end-project with open source feature stores for either real time series or recommendation system. No tutorial on yt is focusing on end-to-end projects with feature stores and real time data
Great to hear that video was helpful to you. Thank you for suggestion about feature stores. Will definitely add feature store to one of my videos. Time series is already on my list. But not sure about Recommendations Systems though. Do you use them in your daily job?
Thank you for the great summary! Are the data imputation and scaling done separately for the training and testing data based on their respective values? Should the imputation and scaling of the testing data be based on the training data?
Glad to hear it was useful! So what you do always is to fit imputation and scaling to the Training and use apply that to the Training. So everything based on Training data idd. Test data needs to be treated as Unseen at all time.
@@maryammiradi Thank you for the clarification! It makes sense to treat the testing data as unseen data and fit the imputation and scaling to the training data and apply the transformation parameters from the training data to both the training and testing data. Great content :-)
Great to hear! From R to Python is easy transition. Here you go you can find my free guide: www.maryammiradi.com/free-guide. And if you want to learn python, this tutorials can help you: ru-vid.com/group/PLzMcBGfZo4-mFu00qxl0a67RhjjZj3jXm&si=Q74zDfz2T2jUJC-8
Hey I have a masters in data science from Italy I’m confuse whether I should join the jobs market or else I’ll try for PhD in Norway due to stipend and then join the Industry… what are you prospective on that as PhD in Europe is only 3 year shorter than USA and we are getting 2500 euros -3000(rest of EU)to 3500 euro stipend in Norway Would you like to do a video on this topic
Sure! I will put this on my list of videos. About your question either you should join the job market or PhD is totally dependent on your character and your purpose.
@@maryammiradi I’m from a middle class family so money is important to me so was thinking of jobs but my thoughts are if I don’t do PhD after my masters maybe I’ll not be able to do it in older age. Also my reasoning was I get an edge in competitive Job market like quant researcher ,trader if I have a PhD vs having a data science masters degree from Italy