AutoML3 :: AutoML for Lifelong Machine Learning Forum

Go back to competition Back to thread list Post in this thread

> How can get time budget of current batch in model.py ?

I cannot get the data info(time budget in public info) in the model.py.
How to get the data info? (As I know, I cannot modify the ingestion.py)

Posted by: Fong @ Aug. 9, 2018, 11:59 a.m.

Hi,

The time info is available in the public_info files available in the input data directory. You can use as guidance the code in the ingestion program to have access to the public_info files. Although you cannot modify the ingestion program, you surely can have access to everything within the input_data directory.

Best

organizers

Posted by: hugo.jair @ Aug. 9, 2018, 12:29 p.m.

Hello
Thank you for your reply.
I can get the public info from the input_data directory, but in the AutoML phase, I'm not sure the input_data filename are the same as that like "AA, B, C, D, E" in feedback phase.
Maybe in the ingestion.py, the code "M = Model()" change to "M = Model(D.info)". This would be great.
Best

Posted by: Fong @ Aug. 9, 2018, 1:31 p.m.

Hi

There is no need to know what the dataset names are, check the code in the ingestion program. Changes cannot be done at this time

Best

Posted by: hugo.jair @ Aug. 9, 2018, 4:30 p.m.

Hi
Thank you for your reply.
Can I assume the default_input_dir name "AutoML3_sample_data" remain the same in the AutoML phase?
(At least I need to know from which directory I can get the data info)
Best

Posted by: Fong @ Aug. 10, 2018, 1:40 a.m.

Hi again,

The directory for the second phase will be AutoML3_input_data_2, I hope this helps.

best

Posted by: hugo.jair @ Aug. 10, 2018, 1:56 a.m.

Hi
Thank you for your reply.
Best

Posted by: Fong @ Aug. 10, 2018, 2:01 a.m.

As you said, we can read the public_info files available in the input data directory. But how do we know which dataset to read the public_info for? Since the basename is not passed to Model.fit() in ingestion.py, we can't know which dataset is being worked upon and which public_info to read. So we can't know how much time we have for this dataset, right? Or did I miss something?

Posted by: aad_freiburg @ Aug. 24, 2018, 1:26 p.m.

Hi,

You could, for instance, read the contents of the AutoML3_input_data_2 folder, this would list the folders (which correspond to the basenames), perhaps I am missing something in your question/comment too?.

Best

Posted by: hugo.jair @ Aug. 24, 2018, 11:15 p.m.

But we can't know which dataset is being worked upon currently. Refer to the loop in line 202. It has a current basename, which is the dataset being worked upon. And then inside this loop, train and test data are passed to M.fit(), but M has no knowledge of the current basename, so how will it know which dataset among the list of datasets to read the public_info for?

Posted by: aad_freiburg @ Aug. 27, 2018, 11:21 a.m.

You will need to pass the basename to M.fit(). This would also solve the issue of reading feature info from the public_info files from https://competitions.codalab.org/forums/16559/2643/.

Posted by: aad_freiburg @ Aug. 27, 2018, 11:24 a.m.


I see, we will work on a solution for this. Passing the name or other info is not that simple ( it will make all existing submissions to fail). We will update all participants when this is addressed, thank you for clarifying the problem. Our apologies for the inconveniences!

Best

Posted by: hugo.jair @ Aug. 29, 2018, 5:52 a.m.
Post in this thread