Refit after one new day

match flagged ← All linear growth benchmarks

The same refit after one new day: a fit of the first 255 days starts the fit of all 256.

Input

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

Output

Largest difference: 0.0012 (0.0012% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	107.7959	107.7956	0.00023
2020-09-14	107.5632	107.5630	0.00024
2020-09-15	108.3426	108.3423	0.00024
2020-09-16	109.9334	109.9330	0.00034
2020-09-17	111.5461	111.5458	0.00031
2020-09-18	112.3360	112.3357	0.00031
2020-09-19	112.0779	112.0775	0.00034
2020-09-20	111.3663	111.3660	0.00034
2020-09-21	111.1337	111.1334	0.00034
2020-09-22	111.9130	111.9127	0.00034
2020-09-23	113.5038	113.5034	0.00044
2020-09-24	115.1165	115.1161	0.00041
2020-09-25	115.9064	115.9060	0.00041
2020-09-26	115.6483	115.6479	0.00044
2020-09-27	114.9368	114.9363	0.00044
2020-09-28	114.7042	114.7037	0.00045
2020-09-29	115.4835	115.4830	0.00044
2020-09-30	117.0743	117.0738	0.00055
2020-10-01	118.6870	118.6865	0.00052
2020-10-02	119.4769	119.4764	0.00052
2020-10-03	119.2188	119.2183	0.00054
2020-10-04	118.5072	118.5067	0.00055
2020-10-05	118.2746	118.2741	0.00055
2020-10-06	119.0539	119.0534	0.00055
2020-10-07	120.6448	120.6441	0.00065
2020-10-08	122.2575	122.2568	0.00062
2020-10-09	123.0473	123.0467	0.00062
2020-10-10	122.7893	122.7886	0.00065
2020-10-11	122.0777	122.0771	0.00065
2020-10-12	121.8451	121.8444	0.00065
2020-10-13	122.6244	122.6237	0.00065
2020-10-14	124.2152	124.2145	0.00075
2020-10-15	125.8279	125.8272	0.00072
2020-10-16	126.6178	126.6171	0.00073
2020-10-17	126.3597	126.3590	0.00075
2020-10-18	125.6482	125.6474	0.00075
2020-10-19	125.4155	125.4148	0.00076
2020-10-20	126.1949	126.1941	0.00076
2020-10-21	127.7857	127.7848	0.00086
2020-10-22	129.3984	129.3975	0.00083
2020-10-23	130.1883	130.1874	0.00083
2020-10-24	129.9302	129.9293	0.00085
2020-10-25	129.2186	129.2178	0.00086
2020-10-26	128.9860	128.9852	0.00086
2020-10-27	129.7653	129.7645	0.00086
2020-10-28	131.3561	131.3552	0.00096
2020-10-29	132.9688	132.9679	0.00093
2020-10-30	133.7587	133.7578	0.00093
2020-10-31	133.5006	133.4997	0.00096
2020-11-01	132.7891	132.7881	0.00096
2020-11-02	132.5565	132.5555	0.00096
2020-11-03	133.3358	133.3348	0.00096
2020-11-04	134.9266	134.9255	0.0011
2020-11-05	136.5393	136.5383	0.0010
2020-11-06	137.3292	137.3282	0.0010
2020-11-07	137.0711	137.0700	0.0011
2020-11-08	136.3595	136.3585	0.0011
2020-11-09	136.1269	136.1259	0.0011
2020-11-10	136.9062	136.9052	0.0011
2020-11-11	138.4971	138.4959	0.0012
2020-11-12	140.1098	140.1086	0.0011
2020-11-13	140.8997	140.8985	0.0011
2020-11-14	140.6416	140.6404	0.0012
2020-11-15	139.9300	139.9288	0.0012

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
Fitting objective (lower is a better fit)	-1762.0385	-1762.2584	0.01
Stationarity residual (lower is closer to the best fit)	29142.8986	15679.3058	0.01

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Fit	26.3 ms	93.2 ms	3.5× faster
Refit from an earlier fit	8.14 ms	21.5 ms	2.6× faster

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-objective",
    "evidenceId": "linear-output-first-v4"
  },
  "fitRequest": "growth-only",
  "configuration": {
    "growth": "linear",
    "scaling": "absmax",
    "seasonalityMode": "additive",
    "holidaysMode": "additive",
    "changepoints": {
      "mode": "auto",
      "count": 25,
      "range": 0.8
    },
    "changepointPriorScale": 0.05,
    "seasonalities": [
      {
        "name": "weekly",
        "periodDays": 7,
        "fourierOrder": 3,
        "priorScale": 10
      }
    ],
    "events": [],
    "regressors": []
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true
  },
  "warmStart": {
    "baseRows": 255,
    "investigateParameterAbsolute": 0.01
  }
}