Skip to content
mape tournament
Español

Run 0204c348a82a · 30 origin dates · 13 products · 9 models · the petrol station

Who Forecast Each Day Best?

Every model forecasts tomorrow’s sales of every product from the days before it. Each tile is that day’s MAPE: lower is better, and the tomato corner marks the winner.

Random Forest won the run with an average MAPE of 24.1, 11.9 points below the naive baseline, Same weekday.

Have sales of your own? Run the same tournament on them — in your browser, and the file never leaves it.

MAPE per model per origin date
Dec 2025–Jan 2026Daysame-weekdayensemble-v2thetaholt-winterssarimaxrandom-forestxgboostprophetlstm
Sun 7
16.9
17.0
32.2
9.7(winner)
10.5
13.6
15.3
16.7
30.6
Mon 8
21.4
12.5
14.5
17.0
15.8
10.0(winner)
13.4
25.6
18.3
Tue 9
21.9
23.7
28.4
19.2
19.6
22.7
34.1
13.4(winner)
31.8
Wed 10
17.9
16.5
21.7
12.1
10.7(winner)
15.7
25.2
18.0
22.9
Thu 11
10.9(winner)
12.8
14.1
13.3
15.1
13.7
19.0
13.5
14.8
Fri 12
6.1(winner)
7.6
9.2
13.3
10.5
9.0
10.2
8.6
9.6
Sat 13
8.5
9.7
23.8
9.7
13.1
12.0
10.5
6.2(winner)
22.1
Sun 14
11.4
17.0
31.9
18.4
15.8
11.4(winner)
14.9
12.1
33.9
Mon 15
76.1
97.8
124.0
60.3
37.8(winner)
51.5
70.0
41.5
135.0
Tue 16
57.8
57.9
65.7
37.9
24.3
17.5(winner)
28.1
26.9
63.3
Wed 17
33.1
34.1
43.6
20.7
42.0
46.3
56.1
20.0(winner)
39.0
Thu 18
12.1
10.7
10.6
14.3
10.0
13.8
14.5
15.8
10.0(winner)
Fri 19
29.9
29.6
29.3
25.7(winner)
26.2
27.1
34.3
27.0
27.5
Sat 20
68.3
50.2
41.3
64.3
55.7
27.3(winner)
29.9
54.5
43.1
Sun 21
10.9
22.4
40.9
10.9
11.7
27.3
24.4
10.8(winner)
40.7
Mon 22
41.5
45.2
54.4
27.5
22.1
19.2(winner)
28.6
26.7
62.4
Tue 23
67.7
62.7
68.9
39.2
27.2(winner)
31.4
36.6
39.8
68.2
Wed 24
13.2(winner)
15.3
16.4
23.0
20.8
14.4
21.8
19.9
16.4
Thu 25
145.1
135.3
129.8
108.4
56.1(winner)
75.0
95.9
88.9
132.1
Fri 26
34.3
24.8
23.2
25.3
17.6(winner)
19.3
27.6
25.7
22.0
Sat 27
46.5
34.2
29.2
35.7
22.6
10.1(winner)
12.8
34.9
31.0
Sun 28
63.0
45.2
40.6
39.2
60.8
15.0(winner)
24.7
47.6
40.7
Mon 29
59.4
48.6
56.3
25.0
47.4
17.6(winner)
39.7
27.6
59.1
Tue 30
18.1(winner)
22.3
24.1
34.0
19.6
19.0
21.4
26.8
24.7
Wed 31
10.3(winner)
11.4
13.8
17.1
13.1
15.9
25.3
16.8
14.7
Thu 1
45.9
40.5
44.7
32.7(winner)
42.1
53.7
70.4
33.4
49.3
Fri 2
12.7(winner)
18.3
22.5
17.0
17.2
15.2
18.1
17.9
22.1
Sat 3
14.8(winner)
22.8
36.2
17.4
22.3
18.6
19.1
19.5
36.0
Sun 4
52.2
44.1
43.3
50.6
49.8
48.2
50.3
43.1(winner)
48.5
Mon 5
51.9
64.4
81.5
21.1
34.3
30.7
41.7
15.2(winner)
83.2

SW same-weekday · ENS ensemble-v2 · TH theta · HW holt-winters · SAR sarimax · RF random-forest · XGB xgboost · prophet · lstm

Scroll the table for all 30 days — the model row and the day column stay in view.

MAPE %lower is better

  • 0–20
  • 20–25
  • 25–30
  • 30–40
  • 40+
  • ND
  • fail
  • short
  • day winner

Best average

24.1 %

random-forest

Steadiest

σ 15.1

sarimax

Leaderboard · avg MAPElower is better

  1. 01Random Forestrandom-forest24.1σ 15.6 · 30 d
  2. 02SARIMAX + weathersarimax26.4σ 15.1 · 30 d
  3. 03Prophetprophet26.5σ 16.5 · 30 d
  4. 04Holt-Wintersholt-winters28.7σ 20.3 · 30 d
  5. 05XGBoostxgboost31.1σ 19.7 · 30 d
  6. 06Ensemble v2ensemble-v235.2σ 27.6 · 30 d
  7. 07Same weekdaysame-weekday36.0σ 29.3 · 30 d
  8. 08Thetatheta40.5σ 29.0 · 30 d
  9. 09LSTMlstm41.8σ 30.4 · 30 d

Order · run 0204c348a82a11 Oct 2026

Dataset
station-year · seed 11 · 425e…
Window
30 origin days · h = 1
Roster
9 models · 3 light, 6 heavy
Engine
py 3.12.15 · uv.lock b7df…
Result hash
c2c3 5020 5edd