batch_pydata.dim() returns whatever dimension the data should be interpreted. However the data is organized in [N,-1] where N = number of events in one batch.
To avoid confusion we may want to either force data to have dimension same as dim() return shape, or maybe simply [-1].
From discussions on the other threads, I believe this is now considered wanted feature (i.e. no reshape in pydata by the fmwk but should be done by a user as it can cause computation overburden).
batch_pydata.dim() returns whatever dimension the data should be interpreted. However the data is organized in [N,-1] where N = number of events in one batch.
To avoid confusion we may want to either force data to have dimension same as dim() return shape, or maybe simply [-1].