This is a great video. I am just finishing up my degree in computer engineering and was stuck on an assignment until I watched this video. It has great production quality and you provide enough information to the user for them to understand what's actually going on. Thanks for this!
Hi, thank you for your amazing video series! One thing I'm wondering about is the fact that you are using your validation data as the test data. I read that the validation data doesn't influence the model in its learning phase. I just want to ask whether it is fine and formally correct to use your validation data as the test data?
Thanks so much for the video, I didn't know about this, I love the way scikit learn integrates with other frameworks, keep it up! I wish I could took this course in person :)!
Hey Jeff! I hope I get a reply as early as possible because I'm working on a deep learning project. So, my dataset is actually image frames (residing in a separate folder) with corresponding XML annotation files in another folder. I want to train a model for object detection. Can I use K fold cross-validation in such kind of dataset? If yes, how? If not, then what would be the best approach?
Crack! Like a good teacher you teached me the intuition and answered the question.. Good i run my k-fold algorithm but now what?, what kind of decissions could i take with this?. Really thanks ! Because on internet people talk mainly about the algorithm but no about the decisions
k-fold is used for many different things, but usually to give you an idea of how your chosen model will perform on new data that it was not trained on. It can also be used to test out new hyperparameters.
Will the kfoldstratify reduce the problem of imbalanced dataset in case of multi class classification ? Or is it required to perform SMOTE apart from kfoldstratify to over the program of imbalanced dataset ?
I would say that its more like ensemble learning method, because you are splitting folds into data with balanced classes. Its not a method for overcoming imbalance classes, for that you can try one of: oversampling, undersampling, adaboost.
![[#2024MAMA] BIGBANG (빅뱅) - 뱅뱅뱅 (BANG BANG BANG) + FANTASTIC BABY | Mnet 241123 방송](http://i.ytimg.com/vi/EQsYfiPR9uM/mqdefault.jpg)
This is a great video. I am just finishing up my degree in computer engineering and was stuck on an assignment until I watched this video. It has great production quality and you provide enough information to the user for them to understand what's actually going on. Thanks for this!
Thank you! This was explained so well and clear. I was looking for a solution all over the internet and finally found one! lets get coding...
Hi, thank you for your amazing video series!
One thing I'm wondering about is the fact that you are using your validation data as the test data. I read that the validation data doesn't influence the model in its learning phase. I just want to ask whether it is fine and formally correct to use your validation data as the test data?
Thanks so much for the video, I didn't know about this, I love the way scikit learn integrates with other frameworks, keep it up!
I wish I could took this course in person :)!
Thanks, it is actually a very popular course in person, as well. But you really get nearly everything online, except graded assignments.
In the first part of your study, you standardized the income, aspect, save_rate and subscriptions columns. Why did you standardize only these columns?
Hey Jeff! I hope I get a reply as early as possible because I'm working on a deep learning project. So, my dataset is actually image frames (residing in a separate folder) with corresponding XML annotation files in another folder. I want to train a model for object detection. Can I use K fold cross-validation in such kind of dataset? If yes, how? If not, then what would be the best approach?
Crack! Like a good teacher you teached me the intuition and answered the question.. Good i run my k-fold algorithm but now what?, what kind of decissions could i take with this?. Really thanks ! Because on internet people talk mainly about the algorithm but no about the decisions
k-fold is used for many different things, but usually to give you an idea of how your chosen model will perform on new data that it was not trained on. It can also be used to test out new hyperparameters.
Hey Jeff, it would be awesome if you could show how to go about combining the models into a single one rather than voting
Sure, let me think about that.
how do you do the normalization with the holdout- is the X_main used to find the scaling and offset and then that is applied to the holdout?
Hi, in the last example, which model (out of the five models) is used for the holdout data?
He used the last one out of 5-folds, but it seems incorrect to report the results of a model which is trained on a particular fold.
How to compare or draw the loss and val_loss in this case?
Will the kfoldstratify reduce the problem of imbalanced dataset in case of multi class classification ? Or is it required to perform SMOTE apart from kfoldstratify to over the program of imbalanced dataset ?
Overcome *
I would say that its more like ensemble learning method, because you are splitting folds into data with balanced classes. Its not a method for overcoming imbalance classes, for that you can try one of: oversampling, undersampling, adaboost.
what if my data are images?
that is my question too