Discussion between Toros91 and ebrahimi

Feeds

12:46

1

A: What to report in the build model, asses model and evaluate results steps of CRISP-DM?

The way you are trying to present the outcome is pretty good. I cannot say that the following procedure is the standard procedure in my scenario I did something like this: This is how I presented to my managers to make them understand the procedure which I followed. I made a slide for each an...

ebrahimi

Thank you very much. However, I 've been asked to present it with all details. Besides, I would appreciate if you could answer this: datascience.stackexchange.com/questions/32812/… especially the question asked in comment section. @Toros91

Toros91

yeah true, the about is clear right? Do you think anything else needs to be appended?

ebrahimi

I first divide the whole dataset into training and test data set. The training set will be used to tune hyperparameters via k-fold cross-validation. Then, the whole training data set is trained using the best determined parameters in the previous step. Finally, the model is tested on the test data set. So, I decided to report:

parameter settings: hyperparameter tuning via grid search , models: the results (confusion matrix) of whole training data set, model description: the coefficients of logistic regression, model assessment: the learning curve, revised parameter settings:nothing is needed and assessment of data mining : the results (confusion matrix) of test data set. @Toros91

Toros91

This would go in the report right?

ebrahimi

Yes, I am thinking of doing so. Is it right?

Toros91

12:46

yeah it is absolutely fine, if you share all such results, it would be great for the next person to understand better

ebrahimi

As long as it is possible for you, could you please provide an exemplary powerpoint which shows what do you present in it?

Toros91

Sure will try to do that and share it with you.

ebrahimi

Thanks a lot and sorry for this inconvenience. Could you please let me know the class-weight in the sklearn's classifiers e.g., logistic regression is classified into which category? @Toros91

Toros91

I dint understand what you meant in that, why would a ml algorithm me classified into one of the sampling techniques?

ebrahimi

In fact, it is possible to use class-weight to handle imbalanced classes for example, chrisalbon.com/machine_learning/logistic_regression/… However, I don't know how this strategy could be classified in one of those categorization for dealing with imbalanced classes. @Toros91

Toros91

12:47

you are talking with respect to this Hyper parameter right?

class_weight='balanced'

ebrahimi

yes

Toros91

For how class_weight works: It penalizes mistakes in samples of class[i] with class_weight[i] instead of 1. So higher class-weight means you want to put more emphasis on a class. From what you say it seems class 0 is 19 times more frequent than class 1. So you should increase the class_weight of class 1 relative to class 0, say {0:.1, 1:.9}. If the class_weight doesn't sum to 1, it will basically change the regularization parameter.

For how class_weight="auto" works, you can have a look at this discussion. In the dev version you can use class_weight="balanced", which is easier to understan…

you know the way it works right?

ebrahimi

I think it is a kind of cost-sensitive method but I am not sure.

Toros91

it is oversampling technique

falls under 1st scenario

if you check carefully

as it just replicates the minor class to make it balanced

ebrahimi

Thanks a lot. Really, I knew oversampling methods like SMOTE, ADASYN

But I don't know how the mechanism of class-weight is working

In fact, I thought it devotes more penalization to misclassifying a positive (1) class as a negative (0) class, so that's why it may be a cost sensitive approach.

Toros91

12:59

if you understood that is more than enough

got what you are looking for?

ebrahimi

yes.

Toros91

ohk man!
All the Best!
Do let me know if you have any other questions. would love to help!
Do accept the answer if you got what you are looking for!

:)

ebrahimi

However, I don't know why if I use oversampling methods from imblearn package, the result is not as good as when I use class-weight.

Thanks a lot for sharing your time and knowledge.

I upvote your answer.

Best wishes,

Bye

Toros91

hmm because the way it works is different

for oversampling using SMOTE, you can expect better results

the reason is it replicates the rare cases and because of that the balancing of class is better

but the other side of the coin, the synthetic values which it inserts might not be logical

so we need to be careful

ebrahimi

I tried SMOTE but class-weight s better results so, my surprise is due to the fact that if both are oversampling methods why class-weight performs better?

Maybe because the current version of SMOTE which is implemented in imblearn can not handle the categorical variables.

s --> provides

Toros91

13:14

hmmm because the way SMOTE works is good for complex cases and normal over sampling works better for normal cases.

Complex cases by which I mean the data is too sparse

ebrahimi

I am so grateful to you.

Besides, the photos you used in your answer are also beautiful.

Have a nice day (or maybe a good night)

Toros91

that was my presentation to my manager and CTO to show them how I used ML for showing them results

Thank you so much for appreciating

It is night over here

Good Night to you as well

ebrahimi

Sorry could you please see the second answer to this question:

8

Q: What is the difference between sample weight and class weight options in scikit learn?

I have class imbalance problem and want to solve this using cost sensitive learning. under sample and over sample give weights to class to use a modified loss function Question Scikit learn has 2 options called class weights and sample weights. Is sample weight actually doing option 2) a...

it says : class-weight gives weights to class to use a modified loss function

Toros91

13:32

hmmmm yeah it give weight to the minority class so that it can replicate and balance them

ebrahimi

Therefore, as it is indicated it is not oversampling

ThanksGood night

Bye

Transcript for

Discussion between Toros91 and ebrahimi