Models¶

Circuits are trained on synthetic data, and summarised in the table below. See Training for how these models work and are built.

Percentages are rates of convergence starting from parameters multiplied or divided by 5 from real values vs the model initial guess.
'floor' is the median impedance calculation count starting from the true parameters.
'excess' is extra calculations beyond 'floor', median and 90^th percentile are shown.

name	circuit	params	truth x/div 5	ml guess	floor	ml excess med	ml excess p90
`rc`	`R0-(R1,C1)`	3	98.3%	100.0%	22	6	7
`rc_l`	`L0-R0-(R1,C1)`	4	94.8%	100.0%	28	0	9
`rq`	`R0-(R1,CPE1)`	4	88.0%	100.0%	28	0	9
`rq_l`	`L0-R0-(R1,CPE1)`	5	87.4%	100.0%	34	0	11
`two_rc`	`R0-(R1,C1)-(R2,C2)`	5	75.5%	100.0%	44	0	11
`two_rc_l`	`L0-R0-(R1,C1)-(R2,C2)`	6	71.5%	100.0%	53	12	13
`two_rq`	`R0-(R1,CPE1)-(R2,CPE2)`	7	55.1%	99.6%	76	15	62
`two_rq_l`	`L0-R0-(R1,CPE1)-(R2,CPE2)`	8	50.3%	98.8%	86	17	188
`randles`	`R0-(R1-W1,CPE1)`	5	70.8%	99.9%	45	0	11
`sei_randles`	`R0-(R1,CPE1)-(R2-W2,CPE2)`	8	41.6%	99.1%	103	0	52
`sei_randles_wo`	`R0-(R1,CPE1)-(R2-Wo2,CPE2)`	9	26.0%	91.3%	276	16	252

Each circuit's page has more detailed benchmarks.