Four regressors

match flagged ← All linear growth benchmarks

256 daily rows with a yes/no regressor and three numeric ones, each standardized differently.

Input

benchmark/inputs/v1/map-regressors-medium.json: 256 history rows and 64 forecast rows.

Output

Largest difference: 0.0042 (0.0043% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	107.1325	107.1342	0.0017
2020-09-14	107.5505	107.5522	0.0017
2020-09-15	107.9540	107.9558	0.0018
2020-09-16	108.3428	108.3447	0.0019
2020-09-17	107.8253	107.8272	0.0019
2020-09-18	110.0095	110.0115	0.0020
2020-09-19	110.3598	110.3619	0.0021
2020-09-20	110.7007	110.7029	0.0022
2020-09-21	109.6759	109.6779	0.0021
2020-09-22	110.0064	110.0085	0.0021
2020-09-23	110.3365	110.3387	0.0022
2020-09-24	110.6695	110.6718	0.0022
2020-09-25	111.4725	111.4748	0.0023
2020-09-26	111.8221	111.8244	0.0023
2020-09-27	114.0074	114.0099	0.0025
2020-09-28	114.3866	114.3891	0.0025
2020-09-29	115.2481	115.2507	0.0025
2020-09-30	112.4892	112.4917	0.0025
2020-10-01	112.9312	112.9337	0.0025
2020-10-02	113.3969	113.3995	0.0026
2020-10-03	114.3501	114.3527	0.0026
2020-10-04	114.8637	114.8664	0.0027
2020-10-05	115.4003	115.4030	0.0027
2020-10-06	117.7807	117.7836	0.0029
2020-10-07	118.8214	118.8243	0.0029
2020-10-08	119.4154	119.4184	0.0029
2020-10-09	118.2009	118.2038	0.0029
2020-10-10	118.8196	118.8225	0.0029
2020-10-11	119.9091	119.9120	0.0029
2020-10-12	120.5392	120.5422	0.0030
2020-10-13	119.8145	119.8175	0.0030
2020-10-14	120.4431	120.4462	0.0031
2020-10-15	123.3525	123.3557	0.0032
2020-10-16	123.9685	123.9718	0.0032
2020-10-17	124.5751	124.5783	0.0033
2020-10-18	123.3482	123.3514	0.0032
2020-10-19	124.3954	124.3986	0.0032
2020-10-20	124.9670	124.9702	0.0032
2020-10-21	125.5265	125.5298	0.0033
2020-10-22	126.0748	126.0781	0.0033
2020-10-23	127.0762	127.0795	0.0033
2020-10-24	129.4280	129.4315	0.0035
2020-10-25	129.9503	129.9538	0.0035
2020-10-26	129.1120	129.1155	0.0035
2020-10-27	128.2673	128.2707	0.0034
2020-10-28	128.7807	128.7842	0.0035
2020-10-29	129.2951	129.2986	0.0035
2020-10-30	129.8121	129.8157	0.0036
2020-10-31	130.7965	130.8000	0.0036
2020-11-01	131.3223	131.3259	0.0036
2020-11-02	133.6761	133.6799	0.0038
2020-11-03	134.2128	134.2166	0.0038
2020-11-04	135.2179	135.2217	0.0038
2020-11-05	133.9410	133.9448	0.0038
2020-11-06	134.4892	134.4931	0.0038
2020-11-07	135.0382	135.0421	0.0039
2020-11-08	134.6934	134.6973	0.0039
2020-11-09	135.2371	135.2410	0.0039
2020-11-10	135.7742	135.7782	0.0040
2020-11-11	138.1248	138.1290	0.0042
2020-11-12	139.1046	139.1087	0.0042
2020-11-13	139.6068	139.6111	0.0042
2020-11-14	138.2703	138.2745	0.0042
2020-11-15	138.7384	138.7426	0.0042

Internal fit differences

The forecasts match, so this case passes. Both libraries search for the best-fitting parameters and stop when they are close enough. On this input they stopped at slightly different points, so these internal numbers differ by more than the threshold where we take a closer look.

Check	Effect Prophet	Python Prophet	Threshold
Stationarity residual (lower is closer to the best fit)	83.9188	35.2434	0.01

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	118 µs	689 µs	5.8× faster
Fit	3.12 ms	10.5 ms	3.4× faster
Predict	506 µs	3.60 ms	7.1× faster
Fit and predict	2.65 ms	14.1 ms	5.3× faster
Convert, fit and predict	3.64 ms	14.8 ms	4.1× faster
First forecast in a new process	179 ms	1.46 s	8.2× faster
Save model to JSON	23 µs	737 µs	32× faster
Load model from JSON	111 µs	3.08 ms	28× faster
Load and predict in a new process	169 ms	1.46 s	8.6× faster

Settings

The case declaration both libraries ran.

{
  "kind": "linear-map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "linear-output-first-v4"
  },
  "configuration": {
    "effectMap": "explicit",
    "changepoints": {
      "mode": "explicit",
      "timestamps": [
        "2020-03-21T00:00:00.000Z",
        "2020-06-29T00:00:00.000Z"
      ]
    },
    "changepointPriorScale": 0.2,
    "seasonalities": [],
    "events": [],
    "regressors": [
      {
        "name": "binary-auto",
        "priorScale": 10,
        "standardization": "auto"
      },
      {
        "name": "numeric-auto",
        "priorScale": 10,
        "standardization": "auto"
      },
      {
        "name": "numeric-always",
        "priorScale": 7,
        "standardization": "always"
      },
      {
        "name": "numeric-never",
        "priorScale": 5,
        "standardization": "never"
      }
    ]
  },
  "effectOptimizer": {
    "algorithm": "auto",
    "maxIterations": 10000,
    "fallback": "newton",
    "lbfgs": {
      "historySize": 5,
      "initAlpha": 0.001,
      "tolObj": 1e-12,
      "tolRelObj": 10000,
      "tolGrad": 1e-8,
      "tolRelGrad": 10000000,
      "tolParam": 1e-8
    }
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true,
    "sigFigs": 12
  }
}