Hello, I encountered an issue when using a dataset stored in TFRecord format: the episodeid is not unique. I intended to use this key as a unique index, but it caused issues. Upon reviewing feature.json, I understand this ID represents the episode ID within a single file. However, multiple episode within a TFRecord file may share the same ID. Is the convert process treating out.npy as a “file” and counting episodes within it? If so, adding a comment would be helpful to prevent similar issues for future users.
Hello, I encountered an issue when using a dataset stored in TFRecord format: the episodeid is not unique. I intended to use this key as a unique index, but it caused issues. Upon reviewing feature.json, I understand this ID represents the episode ID within a single file. However, multiple episode within a TFRecord file may share the same ID. Is the convert process treating out.npy as a “file” and counting episodes within it? If so, adding a comment would be helpful to prevent similar issues for future users.