Schema Mismatch with some parquet files

#10
by jtourille - opened

Hello,

Thanks for this amazing dataset 🤩

I am using snapshot_download and from_parquet for downloading and using the dataset. I get a lot of schema mismatches (450 files), so I add to filter them before loading. Is that expected ?

I can also provide the full list of problematic files.

Best,

Julien

    _ = snapshot_download(
        "PleIAs/common_corpus",
        repo_type="dataset",
        local_dir="./common_corpus",
        max_workers=16,
    )

# ...

ds = Dataset.from_parquet(parquet_files, num_proc=num_proc)

Some information about the schemas. There is an extra field (__index_level_0__) , date is stored as double instead of int64 and there is a missing curator field.

Majority schema:
identifier: string
collection: string
open_type: string
curator: string
license: string
date: int64
title: string
creator: string
language: string
language_type: string
word_count: int64
token_count: int64
text: string

Mismatched schema (450 files):
identifier: string
collection: string
open_type: string
license: string
date: double
title: string
creator: string
language: string
language_type: string
word_count: int64
token_count: int64
text: string
__index_level_0__: int64

Sign up or log in to comment