A regressor forecast by its own sampled model

match ← All linear growth benchmarks

The same rows and regressors, with the regressor's own model sampled too: predictor: { mcmc } against regressor_predictor={"mcmc_samples": 300}, with the main model's seed and 4 chains, which Python passes on from fit. Its 600 draws give 1,200 simulated values, one for each of the main model's 1,200 futures.

Input

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

Output

Largest difference: 0.018 (0.019% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	107.1914	107.1875	0.0039
2020-09-14	107.6416	107.6367	0.0049
2020-09-15	108.0900	108.0832	0.0069
2020-09-16	108.5383	108.5311	0.0072
2020-09-17	107.9404	107.9338	0.0066
2020-09-18	110.1909	110.1834	0.0075
2020-09-19	110.6399	110.6308	0.0091
2020-09-20	111.0890	111.0789	0.010
2020-09-21	110.1006	110.0908	0.0098
2020-09-22	110.5574	110.5457	0.012
2020-09-23	111.0185	111.0066	0.012
2020-09-24	111.4826	111.4698	0.013
2020-09-25	112.3122	112.3000	0.012
2020-09-26	112.7869	112.7732	0.014
2020-09-27	115.0657	115.0522	0.014
2020-09-28	115.5529	115.5386	0.014
2020-09-29	116.4039	116.3902	0.014
2020-09-30	113.7021	113.6867	0.015
2020-10-01	114.2085	114.1926	0.016
2020-10-02	114.7252	114.7078	0.017
2020-10-03	115.6042	115.5880	0.016
2020-10-04	116.1313	116.1145	0.017
2020-10-05	116.6681	116.6508	0.017
2020-10-06	119.0120	118.9943	0.018
2020-10-07	119.9212	119.9061	0.015
2020-10-08	120.4784	120.4631	0.015
2020-10-09	119.2452	119.2278	0.017
2020-10-10	119.8161	119.7979	0.018
2020-10-11	120.7491	120.7329	0.016
2020-10-12	121.3325	121.3161	0.016
2020-10-13	120.5183	120.5002	0.018
2020-10-14	121.1120	121.0944	0.018
2020-10-15	123.8664	123.8521	0.014
2020-10-16	124.4697	124.4545	0.015
2020-10-17	125.0734	125.0576	0.016
2020-10-18	123.8778	123.8610	0.017
2020-10-19	124.8442	124.8297	0.015
2020-10-20	125.4537	125.4382	0.016
2020-10-21	126.0643	126.0495	0.015
2020-10-22	126.6734	126.6587	0.015
2020-10-23	127.6424	127.6291	0.013
2020-10-24	130.0497	130.0367	0.013
2020-10-25	130.6531	130.6400	0.013
2020-10-26	129.8536	129.8398	0.014
2020-10-27	129.0103	128.9968	0.014
2020-10-28	129.6060	129.5931	0.013
2020-10-29	130.1959	130.1830	0.013
2020-10-30	130.7837	130.7698	0.014
2020-10-31	131.7212	131.7088	0.012
2020-11-01	132.2934	132.2807	0.013
2020-11-02	134.6617	134.6497	0.012
2020-11-03	135.2229	135.2097	0.013
2020-11-04	136.1355	136.1249	0.011
2020-11-05	134.8818	134.8699	0.012
2020-11-06	135.4240	135.4109	0.013
2020-11-07	135.9563	135.9420	0.014
2020-11-08	135.4349	135.4216	0.013
2020-11-09	135.9530	135.9392	0.014
2020-11-10	136.4636	136.4482	0.015
2020-11-11	138.7678	138.7534	0.014
2020-11-12	139.6213	139.6085	0.013
2020-11-13	140.1135	140.0991	0.014
2020-11-14	138.7966	138.7799	0.017
2020-11-15	139.2723	139.2547	0.018

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Fit	5.02 s	4.42 s	1.1× slower
Predict	1.68 ms	113 ms	Not compared: different work
Simulate uncertainty	16.3 ms	500 ms	Not compared: different work

Some steps aren't compared, because the two libraries' public calls produce different outputs. Their times show what each call costs, not which library is faster.

• Predict: Python's predict on a posterior also computes every component's range over the draws, while Effect Prophet's predict returns point forecasts and leaves ranges to predictWithIntervals.
• Simulate uncertainty: Python's predict(vectorized=False) also builds the point forecast and every component, while Effect Prophet's predictUncertainty returns only the ranges.

Settings

The case declaration both libraries ran.

{
  "kind": "map",
  "comparison": {
    "kind": "equivalent-posterior",
    "evidenceId": "posterior-nuts-v1"
  },
  "configuration": {
    "growth": "linear",
    "scaling": "absmax",
    "seasonalityMode": "additive",
    "holidaysMode": "additive",
    "changepoints": {
      "mode": "auto",
      "count": 25,
      "range": 0.8
    },
    "changepointPriorScale": 0.05,
    "seasonalities": [],
    "events": [],
    "regressors": [
      {
        "name": "binary-auto",
        "priorScale": 10,
        "standardization": "auto"
      },
      {
        "name": "numeric-auto",
        "priorScale": 10,
        "standardization": "auto",
        "predictor": {
          "mcmcSamples": 300
        }
      },
      {
        "name": "numeric-always",
        "priorScale": 7,
        "standardization": "always"
      },
      {
        "name": "numeric-never",
        "priorScale": 5,
        "standardization": "never"
      }
    ]
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true,
    "sigFigs": 12
  },
  "uncertainty": {
    "seed": 19,
    "samples": 1000,
    "intervalWidth": 0.8,
    "output": "intervals",
    "algorithm": "scalar-continuous-time",
    "evidence": "docs/validation/uncertainty.md#evidence-classes",
    "samplerMemoryLimitBytes": 67108864
  },
  "mcmc": {
    "samples": 300,
    "seed": 19,
    "pointWidthFraction": 0.15
  }
}