Hi,
Could you please explain what a sample is in the submission?
Should we extract features for each smartwatch track or just for each patient (or each patient at a visit, but for the simplicity assume that there is only one visit per patient)?
I mean if there are 10 patients in the test data and each has 5 tracks (recording different tasks), then what is the number of predictions? (10 or 50?)
Thanks,
Balint
Created by Balint Armin Pataki patbaa Hi:
The test data was meant to be released and is now. See syn10701954 Our models assume that for each relevant task to a subchallenge, every record has the same features calculated for it. So if we have a test set, and filter by records which have a task relevant to [SubChallenge 2.1 Upper Limb Action Tremor](https://www.synapse.org/#!Synapse:syn8717496/wiki/448356), then we need to calculate the same features for each of those records.
But it isn't necessary to use the same feature set across different subchallenges. E.g., you may submit 50 features for Upper Limb Action Tremor but only 40 for Upper Limb Resting Dyskinesia. I believe you will need to extract the same features for all tasks for a given subchallenge, since all tasks are scored together. @phil, correct me if I'm wrong.
It looks like the test data are not public yet. Sorry for the confusion.
Solly Thanks for the clarification. I do not see the test data unfortunately, only the train.
So then should we provide the same features for all samples?
For example:
If I extract a feature which is let's say the max acceleration. It works almost perfectly for one task (finger-to-nose), so I don't want to add more features here because many other would worsen the score.
There is an other task (drinking) I want to extract 100 features for it.
Then should I add the 100 features to the finger-to-nose as well? Or is it completely fine because they are separated anyway in the evaluation?
Thanks,
Balint Since each subject's activity replication and side (right vs left) have a separate score, you'll submit a feature vector for each one separately. In the [training data](https://www.synapse.org/#!Synapse:syn10495809/tables/) that will be every row with a score, and in the [test data](https://www.synapse.org/#!Synapse:syn10701954/tables/) that will be every row with a blank value. Note that the set of tasks differs in the 3 subchallenges.
Solly
Drop files to upload
L-Dopa - sample is a task or a patient page is loading…