parquet pipeline #649
parquet pipeline #649
Conversation
dhensle
left a comment
dhensle
left a comment
There was a problem hiding this comment.
Generally looks good and seems to work as designed for me. Wondering if the default pipeline_file_name setting should change in config.py to remove the ".h5". (Had to do some digging here for the actual setting to get this feature turned on!). Note that I think changing the default setting will also use this parquet functionality in the pipeline CI testing, which I think is desirable. Based on the getting_started.ipynb notebooks, I think the .h5 version should be specified explicitly in the settings for the prototype_mtc example.
| If the pipeline_file_name setting ends in ".h5", then the pandas | ||
| HDFStore file format is used, otherwise pipeline files are stored | ||
| as parquet files organized in regular file system directories. |
There was a problem hiding this comment.
I assume documentation on this setting will to be addressed in the other Pydantic task?
| although when using the parquet storage format this file is stored as "None.parquet" | ||
| to maintain a simple consistent file directory structure. | ||
| If the |
There was a problem hiding this comment.
Looks like unfinished thought here...
| store.joinpath(table_name).mkdir(parents=True, exist_ok=True) | ||
| df.to_parquet(store.joinpath(table_name, f"{checkpoint_name}.parquet")) | ||
| else: | ||
| complib = config.setting("pipeline_complib", None) |
There was a problem hiding this comment.
Another setting to not get lost in the Pydantic task.
|
This PR was superseded by #654 |
2 participants
Initial implementation for #645