How can we get the dataset?

[Mayar2009]

Can I get Just part from this data set cause I do not have enough space on my disk?

[kyleclo]

Hey @Mayar2009 yes, I recommend doing something like this:

def process_batch(batch: dict):
    # this downloads both the metadata & full text files for a particular shard
    cmd = ["wget", "-O", batch['input_metadata_path'], batch['input_metadata_url']]
    subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=False)

    cmd = ["wget", "-O", batch['input_pdf_parses_path'], batch['input_pdf_parses_url']]
    subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=False)

    # first, let's filter metadata JSONL to only papers with a particular field of study.
    # we also want to remember which paper IDs to keep, so that we can get their full text later.
    paper_ids_to_keep = set()
    with gzip.open(batch['input_metadata_path'], 'rb') as gz, open(batch['output_metadata_path'], 'wb') as f_out:
        f = io.BufferedReader(gz)
        for line in tqdm(f.readlines()):
            metadata_dict = json.loads(line)
            paper_id = metadata_dict['paper_id']
            mag_field_of_study = metadata_dict['mag_field_of_study']
            if mag_field_of_study and 'Medicine' in mag_field_of_study:     # TODO: <<< change this to your filter
                paper_ids_to_keep.add(paper_id)
                f_out.write(line)

    # now, we get those papers' full text
    with gzip.open(batch['input_pdf_parses_path'], 'rb') as gz, open(batch['output_pdf_parses_path'], 'wb') as f_out:
        f = io.BufferedReader(gz)
        for line in tqdm(f.readlines()):
            metadata_dict = json.loads(line)
            paper_id = metadata_dict['paper_id']
            if paper_id in paper_ids_to_keep:
                f_out.write(line)

    # now delete the raw files to clear up space for other shards
    os.remove(batch['input_metadata_path'])
    os.remove(batch['input_pdf_parses_path'])

and running this on each shard of the dataset (i.e. incrementally downloading & keeping only the parts that you want)

[muguruzawang]

Hey @Mayar2009 yes, I recommend doing something like this:

def process_batch(batch: dict):
    # this downloads both the metadata & full text files for a particular shard
    cmd = ["wget", "-O", batch['input_metadata_path'], batch['input_metadata_url']]
    subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=False)

    cmd = ["wget", "-O", batch['input_pdf_parses_path'], batch['input_pdf_parses_url']]
    subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=False)

    # first, let's filter metadata JSONL to only papers with a particular field of study.
    # we also want to remember which paper IDs to keep, so that we can get their full text later.
    paper_ids_to_keep = set()
    with gzip.open(batch['input_metadata_path'], 'rb') as gz, open(batch['output_metadata_path'], 'wb') as f_out:
        f = io.BufferedReader(gz)
        for line in tqdm(f.readlines()):
            metadata_dict = json.loads(line)
            paper_id = metadata_dict['paper_id']
            mag_field_of_study = metadata_dict['mag_field_of_study']
            if mag_field_of_study and 'Medicine' in mag_field_of_study:     # TODO: <<< change this to your filter
                paper_ids_to_keep.add(paper_id)
                f_out.write(line)

    # now, we get those papers' full text
    with gzip.open(batch['input_pdf_parses_path'], 'rb') as gz, open(batch['output_pdf_parses_path'], 'wb') as f_out:
        f = io.BufferedReader(gz)
        for line in tqdm(f.readlines()):
            metadata_dict = json.loads(line)
            paper_id = metadata_dict['paper_id']
            if paper_id in paper_ids_to_keep:
                f_out.write(line)

    # now delete the raw files to clear up space for other shards
    os.remove(batch['input_metadata_path'])
    os.remove(batch['input_pdf_parses_path'])

and running this on each shard of the dataset (i.e. incrementally downloading & keeping only the parts that you want)

sorry, I run this code but encounter the problem that "FileNotFoundError: [Errno 2] No such file or directory: '/20200705v1/metadata/raw/metadata_0.jsonl.gz'". It seems that the subprocess didn't work well. Do you know the reason?

[smcgrogan]

How do we implement this function? My colab session crashes every time I just try to read the url response.

[kyleclo]

hi @smcgrogan! This downloading function is out of date as we've migrated to releasing S2ORC under the Semantic Scholar API. The README instructions has more up-to-date example of how to do it. Please check that out?

https://github.com/allenai/s2orc?tab=readme-ov-file#download-instructions