improve error messages

[pavelzw]

#74 (comment)

[YYYasin19]

looks good! yeah, we probably should've kept the checks more tight.

[github-actions]

(benchmark 5853999107 / attempt 1)
Base results / Our results / Change

Model	Size	Dump Time	Load Time
sklearn rf 20M	23.4 MiB / 3.0 MiB / 7.67 x	0.02 s / 0.04 s / 2.45 x	0.01 s / 0.04 s / 2.91 x
sklearn rf 20M lzma	7.6 MiB / 2.0 MiB / 3.78 x	14.71 s / 1.23 s / 0.08 x	0.68 s / 0.22 s / 0.33 x
sklearn rf 200M	238.8 MiB / 30.8 MiB / 7.75 x	0.13 s / 0.33 s / 2.51 x	0.18 s / 0.46 s / 2.53 x
sklearn rf 200M lzma	48.4 MiB / 14.8 MiB / 3.26 x	115.10 s / 18.88 s / 0.16 x	4.79 s / 1.73 s / 0.36 x
sklearn rf 1G	1302.2 MiB / 168.0 MiB / 7.75 x	1.08 s / 1.73 s / 1.60 x	1.04 s / 2.22 s / 2.13 x
sklearn rf 1G lzma	295.9 MiB / 99.2 MiB / 2.98 x	725.83 s / 113.36 s / 0.16 x	27.79 s / 10.22 s / 0.37 x
sklearn gb 2M	2.5 MiB / 1.1 MiB / 2.16 x	0.03 s / 0.28 s / 8.87 x	0.04 s / 0.25 s / 6.96 x
sklearn gb 2M lzma	0.7 MiB / 0.2 MiB / 4.42 x	1.13 s / 0.43 s / 0.38 x	0.10 s / 0.26 s / 2.66 x
lgbm gbdt 2M	2.6 MiB / 1.0 MiB / 2.78 x	0.08 s / 0.25 s / 3.04 x	0.01 s / 0.11 s / 8.43 x
lgbm gbdt 2M lzma	0.9 MiB / 0.5 MiB / 1.90 x	1.49 s / 0.51 s / 0.34 x	0.09 s / 0.15 s / 1.80 x
lgbm gbdt 5M	5.3 MiB / 1.9 MiB / 2.81 x	0.14 s / 0.47 s / 3.41 x	0.02 s / 0.23 s / 9.75 x
lgbm gbdt 5M lzma	1.7 MiB / 0.8 MiB / 1.96 x	3.62 s / 1.06 s / 0.29 x	0.16 s / 0.31 s / 1.90 x
lgbm gbdt 20M	22.7 MiB / 7.6 MiB / 3.00 x	0.56 s / 1.85 s / 3.28 x	0.10 s / 0.99 s / 9.86 x
lgbm gbdt 20M lzma	6.3 MiB / 3.0 MiB / 2.09 x	19.12 s / 4.87 s / 0.25 x	0.63 s / 1.28 s / 2.02 x
lgbm gbdt 100M	101.1 MiB / 33.0 MiB / 3.06 x	2.56 s / 8.36 s / 3.26 x	0.52 s / 36.59 s / 69.98 x
lgbm gbdt 100M lzma	25.6 MiB / 10.6 MiB / 2.41 x	91.80 s / 24.27 s / 0.26 x	2.65 s / 5.18 s / 1.95 x
lgbm rf 10M	10.9 MiB / 3.2 MiB / 3.46 x	0.27 s / 0.57 s / 2.07 x	0.04 s / 0.39 s / 10.10 x
lgbm rf 10M lzma	0.7 MiB / 0.4 MiB / 1.86 x	1.95 s / 0.83 s / 0.43 x	0.12 s / 0.43 s / 3.53 x

[StefanSorgQC]

Looks good, thanks.

Can we (in a separate PR) relax the condition to a warning and leave it to sklearn to raise an error if necessary?

[pavelzw]

Can we (in a separate PR) relax the condition to a warning and leave it to sklearn to raise an error if necessary?

Since we directly access every element of the state after the assertion, a warning wouldn't gain us anything since there would then be another error directly afterwards.