BlueMO / bluemo.py
yifAI's picture
Create bluemo.py
045008b verified
# bluemo.py
import json, datasets
from datasets import Features, Value, Sequence, Split, SplitGenerator
_CITATION = ""
_DESCRIPTION = "BlueMO dataset."
class BlueMO(datasets.GeneratorBasedBuilder):
BUILDER_CONFIGS = [
datasets.BuilderConfig(name="default"),
datasets.BuilderConfig(name="proof"),
datasets.BuilderConfig(name="calculation"),
datasets.BuilderConfig(name="mathtext"),
]
def _info(self):
if self.config.name in ("proof", "calculation", "default"):
feats = Features({
"source_file": Value("string"),
"problem_type": Value("string"),
"problem": Value("string"),
"solution": Value("string"),
"remark": Value("string"),
"figures": Sequence(Value("string")),
})
else: # mathtext
feats = Features({
"source_file": Value("string"),
"text": Value("string"),
"figures": Sequence(Value("string")),
})
return datasets.DatasetInfo(description=_DESCRIPTION, citation=_CITATION, features=feats)
def _split_generators(self, dl_manager):
# match your README configs
if self.config.name == "proof":
paths = dl_manager.iter_files(["processed_dataset/proof"])
elif self.config.name == "calculation":
paths = dl_manager.iter_files(["processed_dataset/calculation"])
elif self.config.name == "mathtext":
paths = dl_manager.iter_files(["processed_dataset/text"])
else: # default = proof + calculation
paths = dl_manager.iter_files(["processed_dataset/proof", "processed_dataset/calculation"])
return [SplitGenerator(name=Split.TRAIN, gen_kwargs={"paths": list(paths)})]
def _generate_examples(self, paths):
for i, p in enumerate(sorted(paths)):
with open(p, "r", encoding="utf-8") as f:
obj = json.load(f)
# normalize keys & types
if "figures" in obj and obj["figures"] is None:
obj["figures"] = []
if "figures" in obj:
obj["figures"] = [str(x) for x in (obj["figures"] or [])]
yield str(i), obj