ikhlestov / rarog Goto Github PK

We shouldn't store experiments results in the default clickhouse database.

Main aim is to replace aggregation merge tree with aggregation view. So we have two tables - with raw data and aggregated one. Thus we can omit case, when last entries to the database are not aggregated.

We should support logging of pytorch arrays

Sometimes it's nice to have an ability to store additional experiments metadata in the database. Such as, but not limited to, configs, author, etc.

It will be nice to have an ability to import already saved experiments from tensorboard logs to rarog database.

We should be able to store arrays with ndim > 1

It's required to have an ability to log tensorflow arrays without previous manipulations from the user side.