Question about the variables ("e_form_2000" and "e_form_500") in "analyze_chgnet.py" file

[Claudia-Hello]

Hi,
I got a problem when I ran the analyze_chgnet.py file. It seem that there are no columns named e_form_per_atom_chgnet_relax_steps_2000 and e_form_per_atom_chgnet_relax_steps_500, then it went wrong when I ran the file analyze_chgnet.py. Corresponding code can be seen as follows:

e_form_2000 = "e_form_per_atom_chgnet_relax_steps_2000"
e_form_500 = "e_form_per_atom_chgnet_relax_steps_500"

min_e_diff = 0.1
# structures with smaller energy after longer relaxation need many steps
df_long = df_chgnet.query(f"{e_form_2000} - {e_form_500} < -{min_e_diff}")
# structures with larger energy after longer relaxation are problematic
df_bad = df_chgnet.query(f"{e_form_2000} - {e_form_500} > {min_e_diff}")
# both combined
df_diff = df_chgnet.query(f"abs({e_form_2000} - {e_form_500}) > {min_e_diff}")

[janosh]

that code is a bit outdated. the columns e_form_(500|2000) can be found in the df_chgnet_v020 dataframe that's loaded right above the snippet you posted.

matbench-discovery/models/chgnet/analyze_chgnet.py

Lines 25 to 35 in 3118330

	df_chgnet = df_chgnet_v030 = pd.read_csv(PRED_FILES.CHGNet)
	df_chgnet_v020 = pd.read_csv(
	f"{module_dir}/2023-03-06-chgnet-0.2.0-wbm-IS2RE.csv.gz", index_col=Key.mat_id
	)
	df_chgnet[Key.formula] = df_wbm[Key.formula]
	e_form_2000 = "e_form_per_atom_chgnet_relax_steps_2000"
	e_form_500 = "e_form_per_atom_chgnet_relax_steps_500"
	min_e_diff = 0.1
	# structures with smaller energy after longer relaxation need many steps

feel free to submit a bug fix PR. that said, this analysis was only to check how much effect the max length of MLFF relaxation trajectory has on the ML final energy. answer: not that much. so you could just remove the failing code.

[Claudia-Hello]

thank you so much!