Defaults, irregular dates

match flagged ← All linear growth benchmarks

192 rows spread unevenly over 228 days with two real trend breaks.

Input

benchmark/inputs/v1/map-irregular-medium.json: 192 history rows and 48 forecast rows.

Output

Largest difference: 0.079 (0.11% of the input's range).

Every forecast row from both libraries.

Date	Effect Prophet	Python Prophet	Difference
2020-08-17	84.4431	84.4157	0.027
2020-08-18	84.7250	84.6951	0.030
2020-08-19	85.2959	85.2629	0.033
2020-08-20	85.8726	85.8420	0.031
2020-08-22	86.6564	86.6283	0.028
2020-08-23	87.0704	87.0455	0.025
2020-08-24	87.4516	87.4178	0.034
2020-08-25	87.7334	87.6971	0.036
2020-08-26	88.3044	88.2649	0.039
2020-08-27	88.8810	88.8440	0.037
2020-08-28	89.1273	89.0906	0.037
2020-08-30	90.0789	90.0475	0.031
2020-08-31	90.4600	90.4198	0.040
2020-09-01	90.7419	90.6991	0.043
2020-09-02	91.3128	91.2670	0.046
2020-09-03	91.8895	91.8461	0.043
2020-09-04	92.1357	92.0926	0.043
2020-09-06	93.0873	93.0496	0.038
2020-09-08	93.7503	93.7012	0.049
2020-09-09	94.3213	94.2690	0.052
2020-09-10	94.8980	94.8481	0.050
2020-09-11	95.1442	95.0947	0.049
2020-09-12	95.6818	95.6344	0.047
2020-09-13	96.0958	96.0516	0.044
2020-09-14	96.4770	96.4239	0.053
2020-09-16	97.3297	97.2710	0.059
2020-09-17	97.9064	97.8501	0.056
2020-09-18	98.1526	98.0967	0.056
2020-09-19	98.6902	98.6365	0.054
2020-09-20	99.1043	99.0536	0.051
2020-09-21	99.4854	99.4259	0.060
2020-09-22	99.7673	99.7052	0.062
2020-09-24	100.9149	100.8522	0.063
2020-09-25	101.1611	101.0987	0.062
2020-09-26	101.6987	101.6385	0.060
2020-09-27	102.1127	102.0557	0.057
2020-09-29	102.7757	102.7073	0.068
2020-09-30	103.3467	103.2751	0.072
2020-10-01	103.9233	103.8542	0.069
2020-10-03	104.7071	104.6405	0.067
2020-10-04	105.1212	105.0577	0.063
2020-10-05	105.5023	105.4300	0.072
2020-10-06	105.7842	105.7093	0.075
2020-10-07	106.3551	106.2771	0.078
2020-10-08	106.9318	106.8562	0.076
2020-10-09	107.1780	107.1028	0.075
2020-10-11	108.1296	108.0597	0.070
2020-10-12	108.5108	108.4320	0.079

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)	-745.4706	-744.8759	0.01

Timing

Median of all measured repetitions.

Step	Effect Prophet	Python Prophet	Effect Prophet is
Convert input rows	30 µs	497 µs	17× faster
Fit	6.83 ms	13.3 ms	2.0× faster
Predict	355 µs	3.12 ms	8.8× faster
Fit and predict	6.45 ms	16.7 ms	2.6× faster
Convert, fit and predict	6.47 ms	17.3 ms	2.7× faster
First forecast in a new process	177 ms	1.44 s	8.2× faster
Save model to JSON	36 µs	525 µs	15× faster
Load model from JSON	142 µs	2.34 ms	17× faster
Load and predict in a new process	159 ms	1.43 s	9.0× 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
  }
}