Ranges, eight months

match flagged ← All logistic growth benchmarks

80% ranges from 128 simulated paths on the 256-day history and 64 forecast days.

Input

benchmark/inputs/v1/logistic-implicit-floor-large.json: 256 history rows and 64 forecast rows.

Output

Largest difference: 2.3e-5 (< 0.0001% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-09-13	90.5159	90.5159	1.9e-6
2020-09-14	90.4227	90.4227	2.3e-5
2020-09-15	90.6228	90.6228	1.8e-5
2020-09-16	91.0502	91.0502	1.7e-5
2020-09-17	91.5247	91.5248	1.5e-5
2020-09-18	91.7626	91.7626	7.1e-6
2020-09-19	91.6316	91.6316	1.7e-7
2020-09-20	91.3572	91.3572	2.5e-6
2020-09-21	91.2640	91.2640	2.3e-5
2020-09-22	91.4640	91.4640	1.8e-5
2020-09-23	91.8913	91.8914	1.7e-5
2020-09-24	92.3658	92.3659	1.6e-5
2020-09-25	92.6036	92.6036	6.7e-6
2020-09-26	92.4726	92.4726	6.1e-7
2020-09-27	92.1982	92.1982	2.9e-6
2020-09-28	92.1049	92.1049	2.2e-5
2020-09-29	92.3049	92.3048	1.7e-5
2020-09-30	92.7322	92.7322	1.8e-5
2020-10-01	93.2066	93.2067	1.6e-5
2020-10-02	93.4444	93.4444	6.3e-6
2020-10-03	93.3133	93.3133	9.4e-7
2020-10-04	93.0389	93.0389	3.2e-6
2020-10-05	92.9456	92.9455	2.2e-5
2020-10-06	93.1455	93.1455	1.7e-5
2020-10-07	93.5728	93.5728	1.8e-5
2020-10-08	94.0472	94.0472	1.7e-5
2020-10-09	94.2850	94.2849	6.1e-6
2020-10-10	94.1538	94.1538	1.2e-6
2020-10-11	93.8794	93.8794	3.5e-6
2020-10-12	93.7861	93.7860	2.2e-5
2020-10-13	93.9860	93.9859	1.7e-5
2020-10-14	94.4132	94.4132	1.8e-5
2020-10-15	94.8876	94.8877	1.7e-5
2020-10-16	95.1254	95.1253	5.9e-6
2020-10-17	94.9942	94.9942	1.4e-6
2020-10-18	94.7197	94.7197	3.6e-6
2020-10-19	94.6264	94.6264	2.2e-5
2020-10-20	94.8263	94.8263	1.7e-5
2020-10-21	95.2535	95.2536	1.8e-5
2020-10-22	95.7279	95.7280	1.7e-5
2020-10-23	95.9656	95.9656	5.7e-6
2020-10-24	95.8345	95.8345	1.5e-6
2020-10-25	95.5600	95.5600	3.8e-6
2020-10-26	95.4667	95.4666	2.1e-5
2020-10-27	95.6665	95.6665	1.6e-5
2020-10-28	96.0938	96.0938	1.9e-5
2020-10-29	96.5682	96.5682	1.7e-5
2020-10-30	96.8059	96.8059	5.6e-6
2020-10-31	96.6747	96.6747	1.6e-6
2020-11-01	96.4002	96.4002	3.9e-6
2020-11-02	96.3068	96.3068	2.1e-5
2020-11-03	96.5067	96.5067	1.6e-5
2020-11-04	96.9339	96.9340	1.9e-5
2020-11-05	97.4083	97.4083	1.7e-5
2020-11-06	97.6460	97.6460	5.5e-6
2020-11-07	97.5148	97.5148	1.7e-6
2020-11-08	97.2403	97.2403	3.9e-6
2020-11-09	97.1470	97.1470	2.1e-5
2020-11-10	97.3468	97.3468	1.6e-5
2020-11-11	97.7741	97.7741	1.9e-5
2020-11-12	98.2484	98.2485	1.7e-5
2020-11-13	98.4861	98.4861	5.5e-6
2020-11-14	98.3549	98.3549	1.8e-6
2020-11-15	98.0804	98.0804	4.0e-6

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)	108.1047	4.4441	0.01

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	58 µs	591 µs	10× faster
Save model to JSON	24 µs	669 µs	27× faster
Load model from JSON	137 µs	2.62 ms	19× faster
Simulate uncertainty	1.55 ms	25.9 ms	Not compared: different work
First simulation in a new process	177 ms	1.48 s	Not compared: different work
Load and simulate in a new process	163 ms	1.45 s	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.

• Simulate uncertainty, First simulation in a new process, Load and simulate in a new process: 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-objective",
    "evidenceId": "logistic-map-v1"
  },
  "configuration": {
    "growth": "logistic",
    "scaling": "minmax",
    "seasonalityMode": "additive",
    "holidaysMode": "additive",
    "changepoints": {
      "mode": "explicit",
      "timestamps": [
        "2020-02-10T00:00:00.000Z"
      ]
    },
    "changepointPriorScale": 0.2,
    "seasonalities": [
      {
        "name": "weekly-custom",
        "periodDays": 7,
        "fourierOrder": 2,
        "priorScale": 10
      }
    ],
    "events": [],
    "regressors": []
  },
  "pythonOptimizer": {
    "algorithm": "Auto",
    "maxIterations": 10000,
    "newtonFallback": true,
    "sigFigs": 12
  },
  "uncertainty": {
    "seed": 19,
    "samples": 128,
    "intervalWidth": 0.8,
    "output": "intervals",
    "algorithm": "scalar-continuous-time",
    "evidence": "docs/validation/uncertainty.md#evidence-classes",
    "samplerMemoryLimitBytes": 67108864
  }
}