Refit after a month of new data

match flagged ← All linear growth benchmarks

fit(history, { map: { warmStart } }) from warmStartParameters of a fit of the first 226 days, against fit(df, init=warm_start_params(m)). 256 days in all, fitted with L-BFGS. Timings compare the refit with a fresh fit.

Input

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

Output

Largest difference: 0.0024 (0.0025% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	107.7958	107.7952	0.00059
2020-09-14	107.5632	107.5626	0.00060
2020-09-15	108.3425	108.3418	0.00063
2020-09-16	109.9333	109.9326	0.00068
2020-09-17	111.5460	111.5453	0.00070
2020-09-18	112.3359	112.3351	0.00072
2020-09-19	112.0777	112.0770	0.00074
2020-09-20	111.3662	111.3654	0.00079
2020-09-21	111.1336	111.1328	0.00080
2020-09-22	111.9129	111.9121	0.00082
2020-09-23	113.5037	113.5028	0.00088
2020-09-24	115.1164	115.1155	0.00089
2020-09-25	115.9063	115.9054	0.00092
2020-09-26	115.6482	115.6472	0.00094
2020-09-27	114.9366	114.9356	0.00099
2020-09-28	114.7040	114.7030	0.00099
2020-09-29	115.4833	115.4823	0.0010
2020-09-30	117.0741	117.0730	0.0011
2020-10-01	118.6868	118.6857	0.0011
2020-10-02	119.4767	119.4756	0.0011
2020-10-03	119.2186	119.2175	0.0011
2020-10-04	118.5071	118.5059	0.0012
2020-10-05	118.2744	118.2732	0.0012
2020-10-06	119.0537	119.0525	0.0012
2020-10-07	120.6445	120.6433	0.0013
2020-10-08	122.2572	122.2560	0.0013
2020-10-09	123.0471	123.0458	0.0013
2020-10-10	122.7890	122.7877	0.0013
2020-10-11	122.0775	122.0761	0.0014
2020-10-12	121.8448	121.8435	0.0014
2020-10-13	122.6242	122.6227	0.0014
2020-10-14	124.2150	124.2135	0.0015
2020-10-15	125.8277	125.8262	0.0015
2020-10-16	126.6176	126.6160	0.0015
2020-10-17	126.3594	126.3579	0.0015
2020-10-18	125.6479	125.6463	0.0016
2020-10-19	125.4153	125.4137	0.0016
2020-10-20	126.1946	126.1930	0.0016
2020-10-21	127.7854	127.7837	0.0017
2020-10-22	129.3981	129.3964	0.0017
2020-10-23	130.1880	130.1863	0.0017
2020-10-24	129.9299	129.9281	0.0017
2020-10-25	129.2183	129.2165	0.0018
2020-10-26	128.9857	128.9839	0.0018
2020-10-27	129.7650	129.7632	0.0018
2020-10-28	131.3558	131.3539	0.0019
2020-10-29	132.9685	132.9666	0.0019
2020-10-30	133.7584	133.7565	0.0019
2020-10-31	133.5003	133.4984	0.0019
2020-11-01	132.7887	132.7868	0.0020
2020-11-02	132.5561	132.5541	0.0020
2020-11-03	133.3354	133.3334	0.0020
2020-11-04	134.9262	134.9242	0.0021
2020-11-05	136.5389	136.5369	0.0021
2020-11-06	137.3288	137.3267	0.0021
2020-11-07	137.0707	137.0686	0.0021
2020-11-08	136.3592	136.3570	0.0022
2020-11-09	136.1265	136.1244	0.0022
2020-11-10	136.9058	136.9036	0.0022
2020-11-11	138.4967	138.4944	0.0023
2020-11-12	140.1094	140.1071	0.0023
2020-11-13	140.8993	140.8969	0.0023
2020-11-14	140.6411	140.6388	0.0023
2020-11-15	139.9296	139.9272	0.0024

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)	-1762.2305	-1762.1516	0.01

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Fit	26.7 ms	88.3 ms	3.3× faster
Refit from an earlier fit	27.3 ms	73.2 ms	2.7× 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": 226,
    "investigateParameterAbsolute": 0.01
  }
}