Materials Project time split dataset - load_data_from_json returns None during debugging (conditionally)

[sgbaird]

Sorted by earliest year of reference, limited to experimental entries with fewer than 52 sites: https://figshare.com/articles/dataset/Materials_Project_Time_Split_Data/19991516

How does this seem in terms of a matminer dataset contribution? See How do I do a time-split of Materials Project entries? e.g. pre-2018 vs. post-2018 and sparks-baird/xtal2png#12 (comment) for additional context. Starting to feel like I'm reinventing the wheel by trying to host it myself.

In my own code, I've been running into a strange issue where if I use:

matminer/matminer/utils/io.py

Lines 89 to 177 in 76a529b

	def load_dataframe_from_json(filename, pbar=True, decode=True):
	"""Load pandas dataframe from a json file.
	Automatically decodes and instantiates pymatgen objects in the dataframe.
	Args:
	filename (str): Path to json file. Can be a compressed file (gz and bz2)
	are supported.
	pbar (bool): If true, shows an ASCII progress bar for loading data from disk.
	decode (bool): If true, will automatically decode objects (slow, convenient).
	If false, will return json representations of the objects (fast, inconvenient).
	Returns:
	(Pandas.DataFrame): A pandas dataframe.
	"""
	# Progress bar for reading file with hook
	pbar1 = tqdm(desc=f"Reading file {filename}", position=0, leave=True, ascii=True, disable=not pbar)
	def is_monty_object(o):
	"""
	Determine if an object can be decoded into json
	by monty.
	Args:
	o (object): An object in dict-form.
	Returns:
	(bool)
	"""
	if isinstance(o, dict) and "@class" in o:
	return True
	else:
	return False
	def pbar_hook(obj):
	"""
	A hook for a pbar reading the raw data from json, not
	using monty decoding to decode the object.
	Args:
	obj (object): A dict-like
	Returns:
	obj (object)
	"""
	if is_monty_object(obj):
	pbar1.update(1)
	sys.stderr.flush()
	return obj
	# Progress bar for decoding objects
	pbar2 = tqdm(desc=f"Decoding objects from {filename}", position=0, leave=True, ascii=True, disable=not pbar)
	class MontyDecoderPbar(MontyDecoder):
	"""
	A pbar-friendly version of MontyDecoder.
	"""
	def process_decoded(self, d):
	if isinstance(d, dict) and "data" in d and "index" in d and "columns" in d:
	# total number of objects to decode
	# is the number of @class mentions
	pbar2.total = str(d).count("@class")
	elif is_monty_object(d):
	pbar2.update(1)
	sys.stderr.flush()
	return super().process_decoded(d)
	if decode:
	decoder = MontyDecoderPbar if pbar else MontyDecoder
	else:
	decoder = None
	hook = pbar_hook if pbar else lambda x: x
	with zopen(filename, "rb") as f:
	dataframe_data = json.load(f, cls=decoder, object_hook=hook)
	pbar1.close()
	pbar2.close()
	# if only keys are data, columns, index then orient=split
	if isinstance(dataframe_data, dict):
	if set(dataframe_data.keys()) == {"data", "columns", "index"}:
	return pandas.DataFrame(**dataframe_data)
	else:
	return pandas.DataFrame(dataframe_data)

It returns None during an uninterrupted debugging run, but if I set a breakpoint and run the line manually in the debug console (VS Code) then it returns the expected DataFrame.
See https://github.com/sparks-baird/mp-time-split/runs/6739787243?check_suite_focus=true/#step:5:1