Ludwig for multivariate time series for regression and classification problem

[ikramelmbarki]

I read the issue #668 for multivariate time series with classification but the data contains only timestamps data as input and it's a classification problem.

My data look like this. I wanna do prediction and create API to visualize the result, is Ludwig will do the following:

1. Data preprocessing for different types.
2. Does it test all available algorithms and gives the best one in a python file, like TPOT do(but it does not work with time series).
3. Could a test different kind of dataset and change hyperparameters.
4. Could I specify the algorithms that I wanna use? if Ludwig can do anything from the following, you can specify, please.

Thank you for your help, I tried a lot of open source packages but it does not work for my case.

[w4nderlust]

@ikramelmbarki sorry from your message I can't really understand what are you trying to do. What are you trying to predict? and using which fields of your data?

Ludwig has a new hyperparameter optimization functionality which is now in a separate branch but will be merged soon, which allows you to train several models with all the different architectures and all the different parameters which looks like your 2, 3 and 4 point. It also already does data preprocessing for many datatypes already, which sounds like your point 1. But until I understand what are you trying to do I can't say for sure if Ludwig is the tool you are looking for.

[ikramelmbarki]

@ikramelmbarki sorry from your message I can't really understand what are you trying to do. What are you trying to predict? and using which fields of your data?

Ludwig has a new hyperparameter optimization functionality which is now in a separate branch but will be merged soon, which allows you to train several models with all the different architectures and all the different parameters which looks like your 2, 3 and 4 point. It also already does data preprocessing for many datatypes already, which sounds like your point 1. But until I understand what are you trying to do I can't say for sure if Ludwig is the tool you are looking for.

I'm trying to predict the continuous value' Montant engagé', and I could have other exogenous data like weather... and I'm wondering if I could use the same thing for classification problem or multiclassification problem

[w4nderlust]

And using which input columns?

[ikramelmbarki]

And using which input columns?

all the other columns except the target value and I could have another exogenous variable like vacancy, weather...

[ikramelmbarki]

Ludwig has a new hyperparameter optimization functionality

could I use this new hyperparameter optimization functionality now?

[w4nderlust]

Ludwig has a new hyperparameter optimization functionality

could I use this new hyperparameter optimization functionality now?

Yes, just pull from the hyperopt branch

[w4nderlust]

And using which input columns?

all the other columns except the target value and I could have another exogenous variable like vacancy, weather...

If the other columns have values belonging to the datatype Ludwig supports, then definitely yes.

[ikramelmbarki]

the column's type are objects or numerical and the date column that I use as index

[w4nderlust]

Object is a pandas datatype, not a Ludwig datatype. You need to figure out what Ludwig datatype you should use.

[ikramelmbarki]

it uses strings and numeric variables, should I transform my pandas' objects?

[w4nderlust]

Strings can be encoded with either category, sequence or text, depending on what information they actually contain.

[w4nderlust]

Closing this due to lack of activity, can reopen if needed.