JSON diff has unreasonable threshold for big diff

[jmarrec]

As stated on NREL/EnergyPlus#8833 (comment)

The json big diffs: I think the epregression script is unreasonable:

EnergyPlusRegressionTool/epregressions/runtests.py

Lines 630 to 643 in 237b8a3

	for col_num in range(num_values):
	num_values_checked += 1
	value_1 = this_row_data_1[col_num]
	value_2 = this_row_data_2[col_num]
	if not value_1 == value_2:
	f_1 = float(value_1)
	f_2 = float(value_2)
	if abs(f_1 - f_2) > 0.00001:
	resulting_diff_type = 'Big Diffs'
	num_big_diffs += 1
	else:
	if 'Big' not in resulting_diff_type:
	resulting_diff_type = 'Small Diffs'
	num_small_diffs += 1

It assumes an absolute difference > 0.0001 is a big diff. This for eg:

It's categorized as a big diff, but in fact it's a 1e-12 difference in %...

In [1]: f_1 = 30310597.103115804

In [2]: f_2 = 30310597.102913737

In [3]: abs(f_1 - f_2) > 0.00001
Out[6]: True

In [7]: (f_1 - f_2) / f_1
Out[7]: 6.6665528125937005e-12