model	mean	median	std	min	max	config_key	nsteps	clusters	subjects
1	VAR_grp_w/o	2.667283	2.311391	1.299086	0.841197	8.663879	nsteps=300_clusters=20_subjects=1	300	20	1
2	VAR_grp_w/	2.130462	1.797232	1.090619	0.826061	7.556795	nsteps=300_clusters=20_subjects=1	300	20	1
3	VAR_ind	2.016944	1.720799	1.039695	0.838965	7.526899	nsteps=300_clusters=20_subjects=1	300	20	1
4	LSTM_grp_w/o	2.552010	2.189134	1.190465	1.140162	9.077024	nsteps=300_clusters=20_subjects=1	300	20	1
5	LSTM_grp_w/	2.399155	2.104400	1.098909	1.107469	7.647618	nsteps=300_clusters=20_subjects=1	300	20	1
...	...	...	...	...	...	...	...	...	...	...
170	VAR_grp_w/	1.556520	1.535209	0.341383	0.839781	2.518587	nsteps=1500_clusters=1_subjects=20	1500	1	20
171	VAR_ind	1.544645	1.519936	0.339314	0.845402	2.562656	nsteps=1500_clusters=1_subjects=20	1500	1	20
172	LSTM_grp_w/o	1.390897	1.377687	0.259057	0.786301	2.041435	nsteps=1500_clusters=1_subjects=20	1500	1	20
173	LSTM_grp_w/	1.390510	1.395509	0.251880	0.840643	2.138441	nsteps=1500_clusters=1_subjects=20	1500	1	20
174	LSTM_ind	1.555531	1.552534	0.307968	0.847127	2.255837	nsteps=1500_clusters=1_subjects=20	1500	1	20

150 rows × 10 columns

mean mse by nsteps for each model

mean mse by clusters for each model

mean mse by clusters, faceted by nsteps

Bad subject count analysis

Pick subject whose lstm_w performance is worse than var_w

	nsteps	clusters	bad_subject_count
0	300	20	17
1	300	10	17
2	300	5	16
3	300	2	10
4	300	1	5
5	600	20	15
6	600	10	11
7	600	5	10
8	600	2	3
9	600	1	0
10	900	20	9
11	900	10	8
12	900	5	3
13	900	2	3
14	900	1	0
15	1200	20	10
16	1200	10	7
17	1200	5	2
18	1200	2	1
19	1200	1	0
20	1500	20	5
21	1500	10	6
22	1500	5	3
23	1500	2	3
24	1500	1	0