I want to know why the classification learning model takes too long to learn.

I had the whole model learn to find the best model in the classifier.

But 2 models took too long and I don't know if this is a feature of the model or just an error. (2nd SVM, 3rd SVM)

The longest learning time among the models was about 2 to 3 days, but the above 2 models have not been completed even after 5 days, so they are currently canceled.

Is this an error or is the model originally taking a long time?

-------

Data (normalized)

Number of rows : 806371

Response variable : 0,1,2,3,4,5,6

Number of Features : 12

Cross-validation : 5

Test Data : 20%

(I inserted the data as above, selected all models, and conducted all training)

(I used the contents with Korean and returned translator.I'm sorry if the content is weird)

Answer 1

[Gagan Agarwal]

Hi 현준 김

The learning time of a model can vary depending on several factors, including the complexity of the data, the size of the dataset, the chosen algorithm etc. It is also not uncommon for certain models to take longer to train compared to others.

To determine whether the extended training time is an error or expected behavior, you can consider investigating the following parameters:

1. Model Complexity: SVM models can be computationally demanding, particularly when working with extensive datasets. Given the large number of rows in your dataset, this could be a contributing factor to the extended training times.
2. Check for Errors: Review the training logs and error messages for the 2nd and 3rd SVM models. Look for any indications of errors or warnings that might explain the extended training time.
3. Optimize Hyperparameters: Experiment with different hyperparameter settings for the SVM models and consider using techniques like grid search or random search to find the optimal combination of hyperparameters that balances accuracy and training time.
4. Sample Data: Since your dataset is extremely large, you could consider sampling a subset of the data to train the models initially. This will help in identifying any issues or errors more quickly without waiting for the entire dataset to be processed.

I hope this helps!