Machine learning in trading: theory, models, practice and algo-trading - page 3625

New comment
 
Evgeni Gavrilovi #:

Interesting. Certainly haven't given up, I'm experimenting with hyperparameters in the catbusters

To get the metrics to match the competition metrics:

model = CatBoostClassifier(iterations=1000,
                               custom_loss=['MultiClass'],
                               eval_metric='Accuracy',
                               auto_class_weights='Balanced',
                               verbose=True,
                               use_best_model=True,
                               task_type='CPU')
model.fit(X_train, y_train, eval_set=(X_test, y_test),
              early_stopping_rounds=50, plot=False)
 
Maxim Dmitrievsky #:

To calculate the correct metrics to match the competition:

I think that in this contest the first place will be taken by a person outside the top 10

This sometimes happens on kaggle, for example here on the private dataset the leader was as high as 2089 in the public ranking!

https://www.kaggle.com/competitions/icr-identify-age-related-conditions/leaderboard

 
Evgeni Gavrilovi #:

I think the top spot in this contest will go to someone outside the top 10

This sometimes happens on kaggle, for example here on the private dataset the leader was as high as 2089th place in the public ranking!

https://www.kaggle.com/competitions/icr-identify-age-related-conditions/leaderboard

Well, getting a high scor above 0.5 in this contest is probably as likely as creating a profitable ts on increments with a single lag :)

 
A relatively new method in kozula is coincidence analysis. I haven't studied it in detail yet, for the future.
 

Hopfield and Hinton were given a Nobel in physics on a wave of hype :)

"Godfather of Artificial Intelligence" also talks about some kind of security. More scaremongering? :)


 

Prado poured in a new laptop for the contest, using Transformers

Mm. Pythorch. Pythorch. I know.


Hello Max,

Exciting news! We've just introduced aninnovative new way to tackle theAdia LabCausal Discovery Challenge. If you're looking for a fresh and powerful approach, we now offer aNeural Network-based solution that could help you push your results further.

You can get started right away with theNeural Network Quickstart Notebook here :

👉Neural Network approach


To be honest, I gave up on the contest because I ran out of simple ideas and no time for complex ones.

slipped from 30th to 70th place out of 6500+ participants.
 

A curious interview, in my opinion. How our IT industry survived.


 
Maxim Dmitrievsky #:
A relatively new method in kozula is coincidence analysis. Haven't studied it in detail yet, for the future.

Another paper

Files:
s13012-020-01070-3.pdf.zip  740 kb
 

The most popular programming languages


https://dzen.ru/video/watch/67116da48f3b99294f789d2c

 
Maxim Dmitrievsky #:

Another paper

Qualitative comparative analysis(QCA) is probably the more correct method.

It looks like a fast search through matrix multiplication. The disadvantage is that it works only with binary data. But it seems to solve the problem through fuzzy sets. Or through one-hot-encoding it is probably possible.

1...361836193620362136223623362436253626362736283629363036313632...3713
New comment