Open jsmeredith opened 10 years ago
"That could be an artifact of the google public data explorer expecting everything to be time based, but there's no reason to keep duplicates in the table."
Yes this is the case. When we set up the Google public data set, you couldn't get any charting without a date column. So the dates were contrived to get that set up. So if we are abandoning Google I'd say migrate to hardware release dates.
The dates in the results.csv are meaningless, but it looks like we may have a (nearly) full set of duplicated in that table. That could be an artifact of the google public data explorer expecting everything to be time based, but there's no reason to keep duplicates in the table.
Also, we should have a different sentinel or empty value for unknown dates.
For other csv's, like specs for a platform/device, if we want to keep the date column, I'd recommend changing their date be the release date of the architecture, or the cuda platform version, etc. It would be meaningful that way. (Otherwise, I'd say remove the date for everything but results.)