rogermt commited on
Commit
54f7e7d
·
verified ·
1 Parent(s): 51a8fd4

Add Kaggle submission instructions and status

Browse files
Files changed (1) hide show
  1. medal-solvers/KAGGLE_SUBMIT.md +50 -0
medal-solvers/KAGGLE_SUBMIT.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Kaggle Submission Instructions
2
+
3
+ ## Current Status: LB 6053.17 (V17), Bronze threshold: 6055.14
4
+
5
+ ## How to Submit (Kaggle Notebook)
6
+
7
+ ```python
8
+ # In a Kaggle notebook with neurogolf-2026 competition data:
9
+
10
+ !pip install onnx onnxruntime numpy
11
+
12
+ # Clone repo
13
+ !git clone https://huggingface.co/rogermt/neurogolf-solver /kaggle/working/neurogolf-solver
14
+
15
+ # Run swap_and_submit
16
+ import os
17
+ os.chdir('/kaggle/working/neurogolf-solver/medal-solvers')
18
+
19
+ !python swap_and_submit.py \
20
+ --base /kaggle/working/neurogolf-solver/submission-6043.zip \
21
+ --task-data-dir /kaggle/input/competitions/neurogolf-2026 \
22
+ --output /kaggle/working/submission.zip
23
+ ```
24
+
25
+ ## What swap_and_submit.py does:
26
+ 1. Opens submission-6043.zip (base submission)
27
+ 2. Auto-discovers ALL .onnx files in `optimized/` folder:
28
+ - `optimized/task255.onnx` (11.570 pts, +4.93 from original)
29
+ - `optimized/task285.onnx` (8.818 pts, +2.83 from original)
30
+ - `optimized/task319.onnx` (11.244 pts, +3.32 from original)
31
+ 3. Validates each model against task data (0 failures required)
32
+ 4. Creates new submission.zip with replaced models
33
+
34
+ ## Score Budget
35
+
36
+ | Task | Original | Optimized | Gain | Status |
37
+ |------|----------|-----------|------|--------|
38
+ | 285 | 5.98 | 8.82 | +2.83 | ✅ Confirmed on Kaggle |
39
+ | 255 | 6.64 | 11.57 | +4.93 | ✅ In submission |
40
+ | 319 | 7.92 | 11.24 | +3.32 | ✅ In submission |
41
+ | **Total** | | | **+11.08** | **LB: 6053.17** |
42
+
43
+ ## Gap to Bronze: 1.97 pts
44
+
45
+ To close the gap, would need to optimize another task model.
46
+ Largest models with most optimization potential:
47
+ - task209.onnx (1.29MB), task366.onnx (1.27MB), task084.onnx (1.13MB)
48
+ - task076.onnx (949KB), task233.onnx (939KB), task153.onnx (862KB)
49
+
50
+ Each would require reverse-engineering the task rule (like we did for 255/285/319).