Commit
·
e4bfb2b
1
Parent(s):
2109e62
add code for using the ds
Browse files- dataset_code.py +99 -0
- example_usage.py +10 -0
dataset_code.py
ADDED
|
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import datasets
|
| 2 |
+
import logging
|
| 3 |
+
import csv
|
| 4 |
+
import sys
|
| 5 |
+
from csv import DictReader
|
| 6 |
+
csv.field_size_limit(sys.maxsize)
|
| 7 |
+
|
| 8 |
+
logger = logging.getLogger(__name__)
|
| 9 |
+
|
| 10 |
+
class FFV4Config(datasets.BuilderConfig):
|
| 11 |
+
"""BuilderConfig for SuperGLUE."""
|
| 12 |
+
|
| 13 |
+
def __init__(self, filename: str, info: str, **kwargs):
|
| 14 |
+
"""BuilderConfig for SuperGLUE.
|
| 15 |
+
|
| 16 |
+
Args:
|
| 17 |
+
features: *list[string]*, list of the features that will appear in the
|
| 18 |
+
feature dict. Should not include "label".
|
| 19 |
+
filename: *string*, csvfile for the dataset.
|
| 20 |
+
info: *string*, for information about the data set.
|
| 21 |
+
**kwargs: keyword arguments forwarded to super.
|
| 22 |
+
"""
|
| 23 |
+
# Version history:
|
| 24 |
+
# 0.0.1: Initial version
|
| 25 |
+
super().__init__(version=datasets.Version("0.0.1"), **kwargs)
|
| 26 |
+
self.filename = filename
|
| 27 |
+
self.info = info
|
| 28 |
+
|
| 29 |
+
class FFV4(datasets.GeneratorBasedBuilder):
|
| 30 |
+
"""The thing"""
|
| 31 |
+
|
| 32 |
+
BUILDER_CONFIGS = [
|
| 33 |
+
FFV4Config(
|
| 34 |
+
name="notebook_defaults",
|
| 35 |
+
filename="notebook_defaults.csv",
|
| 36 |
+
info="the result of using the default values in the V4 ffarchive notebook, except without the TS/RD filter",
|
| 37 |
+
),
|
| 38 |
+
FFV4Config(
|
| 39 |
+
name="notebook_defaults_ratio0.8_likes10",
|
| 40 |
+
filename="ratio0.8_likes10.csv",
|
| 41 |
+
info="default filter, but with the score filter replaced with '.ratio > 0.8, .likes > 10'",
|
| 42 |
+
),
|
| 43 |
+
]
|
| 44 |
+
DEFAULT_CONFIG_NAME = "notebook_defaults"
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
def _info(self):
|
| 48 |
+
return datasets.DatasetInfo(
|
| 49 |
+
description="Garbage datasets for LLM training",
|
| 50 |
+
features=datasets.Features(
|
| 51 |
+
{
|
| 52 |
+
"id": datasets.Value("int32"),
|
| 53 |
+
"header": datasets.Value("string"),
|
| 54 |
+
"story": datasets.Value("string"),
|
| 55 |
+
}
|
| 56 |
+
),
|
| 57 |
+
homepage="https://main.horse",
|
| 58 |
+
)
|
| 59 |
+
|
| 60 |
+
|
| 61 |
+
def _split_generators(self, x):
|
| 62 |
+
return [
|
| 63 |
+
datasets.SplitGenerator('everything', gen_kwargs={"filepath": self.config.filename}),
|
| 64 |
+
]
|
| 65 |
+
'''
|
| 66 |
+
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
|
| 67 |
+
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
|
| 68 |
+
'''
|
| 69 |
+
|
| 70 |
+
def _generate_examples(self, filepath):
|
| 71 |
+
"""This function returns the examples in the raw (text) form."""
|
| 72 |
+
logger.info("generating examples from = %s", filepath)
|
| 73 |
+
with open(filepath, encoding="utf-8") as f:
|
| 74 |
+
dr = DictReader(f)
|
| 75 |
+
for d in dr:
|
| 76 |
+
yield d['id'],d
|
| 77 |
+
'''
|
| 78 |
+
squad = json.load(f)
|
| 79 |
+
for article in squad["data"]:
|
| 80 |
+
title = article.get("title", "")
|
| 81 |
+
for paragraph in article["paragraphs"]:
|
| 82 |
+
context = paragraph["context"] # do not strip leading blank spaces GH-2585
|
| 83 |
+
for qa in paragraph["qas"]:
|
| 84 |
+
answer_starts = [answer["answer_start"] for answer in qa["answers"]]
|
| 85 |
+
answers = [answer["text"] for answer in qa["answers"]]
|
| 86 |
+
# Features currently used are "context", "question", and "answers".
|
| 87 |
+
# Others are extracted here for the ease of future expansions.
|
| 88 |
+
yield key, {
|
| 89 |
+
"title": title,
|
| 90 |
+
"context": context,
|
| 91 |
+
"question": qa["question"],
|
| 92 |
+
"id": qa["id"],
|
| 93 |
+
"answers": {
|
| 94 |
+
"answer_start": answer_starts,
|
| 95 |
+
"text": answers,
|
| 96 |
+
},
|
| 97 |
+
}
|
| 98 |
+
key += 1
|
| 99 |
+
'''
|
example_usage.py
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import datasets
|
| 2 |
+
|
| 3 |
+
ds = datasets.load_dataset('./dataset_code.py', 'notebook_defaults')
|
| 4 |
+
|
| 5 |
+
ds_real = ds['everything'] # there is no such thing as a train/test split here
|
| 6 |
+
one_item = ds_real[0] # grab first story, and truncuate the text of it to first 1000 characters
|
| 7 |
+
one_item_truncuated = one_item | {'story': one_item['story'][:1000]}
|
| 8 |
+
|
| 9 |
+
print(ds)
|
| 10 |
+
print(one_item_truncuated)
|