def _register_arrow_json_readoptions_serializer():
import ray
if (
os.environ.get(
"RAY_DISABLE_CUSTOM_ARROW_JSON_OPTIONS_SERIALIZATION",
"0",
)
== "1"
):
import logging
logger = logging.getLogger(__name__)
logger.info("Disabling custom Arrow JSON ReadOptions serialization.")
return
try:
import pyarrow.json as pajson
except ModuleNotFoundError:
return
ray.util.register_serializer(
pajson.ReadOptions,
serializer=lambda opts: (opts.use_threads, opts.block_size),
deserializer=lambda args: pajson.ReadOptions(*args),
)
def _read_file(self, f: "pyarrow.NativeFile", **arrow_reader_args):
from pyarrow import json
read_options = arrow_reader_args.pop(
"read_options", json.ReadOptions(use_threads=False))
return json.read_json(f,
read_options=read_options,
**arrow_reader_args)
def json_read(read_paths: List[str]):
logger.debug(f"Reading {len(read_paths)} files.")
tables = []
for read_path in read_paths:
with filesystem.open_input_file(read_path) as f:
tables.append(
json.read_json(
f,
read_options=json.ReadOptions(use_threads=False),
**arrow_json_args))
block = ArrowBlock(pa.concat_tables(tables))
return block, block.get_metadata(input_files=read_paths)
class JsonConfig(nlp.BuilderConfig):
"""BuilderConfig for JSON."""
read_options: paj.ReadOptions = paj.ReadOptions()
parse_options: paj.ParseOptions = paj.ParseOptions()
@property
def pa_read_options(self):
return self.read_options
@property
def pa_parse_options(self):
return self.parse_options
def _init_table_from_path(self):
if '.jsonl' in self.path.suffixes:
# Can read ".jsonl" or ".jsonl.gz"
import pyarrow.json as paj
self.table = paj.read_json(
str(self.path),
read_options=paj.ReadOptions(
# magic constants:
# 894 - estimated average number of bytes per JSON item manifest
# 10000 - how many items we want to have in a chunk (Arrow's "batch")
block_size=894 * 10000))
elif '.arrow' == self.path.suffixes[-1]:
# Can read ".arrow"
import pyarrow as pa
mmap = pa.memory_map(str(self.path))
stream = pa.ipc.open_file(mmap)
self.table = stream.read_all()
else:
raise ValueError(f"Unknown LazyDict file format : '{self.path}'")
def pa_read_options(self):
return paj.ReadOptions(use_threads=self.use_threads,
block_size=self.block_size)
def _generate_tables(self, files):
for file_idx, file in enumerate(files):
# If the file is one json object and if we need to look at the list of items in one specific field
if self.config.field is not None:
with open(file, encoding="utf-8") as f:
dataset = json.load(f)
# We keep only the field we are interested in
dataset = dataset[self.config.field]
# We accept two format: a list of dicts or a dict of lists
if isinstance(dataset, (list, tuple)):
mapping = {
col: [dataset[i][col] for i in range(len(dataset))]
for col in dataset[0].keys()
}
else:
mapping = dataset
pa_table = pa.Table.from_pydict(mapping=mapping)
yield file_idx, self._cast_classlabels(pa_table)
# If the file has one json object per line
else:
with open(file, "rb") as f:
batch_idx = 0
# Use block_size equal to the chunk size divided by 32 to leverage multithreading
# Set a default minimum value of 16kB if the chunk size is really small
block_size = max(self.config.chunksize // 32, 16 << 10)
while True:
batch = f.read(self.config.chunksize)
if not batch:
break
# Finish current line
try:
batch += f.readline()
except (AttributeError, io.UnsupportedOperation):
batch += readline(f)
try:
while True:
try:
pa_table = paj.read_json(
io.BytesIO(batch),
read_options=paj.ReadOptions(
block_size=block_size))
break
except (pa.ArrowInvalid,
pa.ArrowNotImplementedError) as e:
if (isinstance(e, pa.ArrowInvalid)
and "straddling" not in str(e)
or block_size > len(batch)):
raise
else:
# Increase the block size in case it was too small.
# The block size will be reset for the next file.
logger.debug(
f"Batch of {len(batch)} bytes couldn't be parsed with block_size={block_size}. Retrying with block_size={block_size * 2}."
)
block_size *= 2
except pa.ArrowInvalid as e:
logger.error(
f"Failed to read file '{file}' with error {type(e)}: {e}"
)
try:
with open(file, encoding="utf-8") as f:
dataset = json.load(f)
except json.JSONDecodeError:
raise e
raise ValueError(
f"Not able to read records in the JSON file at {file}. "
f"You should probably indicate the field of the JSON file containing your records. "
f"This JSON file contain the following fields: {str(list(dataset.keys()))}. "
f"Select the correct one and provide it as `field='XXX'` to the dataset loading method. "
) from None
# Uncomment for debugging (will print the Arrow table size and elements)
# logger.warning(f"pa_table: {pa_table} num rows: {pa_table.num_rows}")
# logger.warning('\n'.join(str(pa_table.slice(i, 1).to_pydict()) for i in range(pa_table.num_rows)))
yield (file_idx,
batch_idx), self._cast_classlabels(pa_table)
batch_idx += 1
