Update README for v3 (306/400)

README.md

@@ -1,215 +1,30 @@
-# NeuroGolf Solver
-
-Builds minimal ONNX networks for ARC-AGI tasks. Currently solves **294/400** on CPU (15s budget). With more time budget on faster hardware, expected .
-
-## Repo
-https://huggingface.co/rogermt/neurogolf-solver
-
-Here’s the **clear, high‑signal overview** of the ARC‑AGI *NeuroGolf Championship* competition
-
----
-
-# 🧠 **What the Competition Actually Is**
-The NeuroGolf Championship is a **model‑building competition** based on the ARC‑AGI benchmark.  
-Your job is to produce **400 ONNX models**, one per task, each mapping:
-
-```
-input:  [1, 10, 30, 30]  one‑hot grid
-output: [1, 10, 30, 30]  one‑hot grid
-```
-
-Each task is a small visual transformation puzzle (flip, rotate, color map, crop, tile, etc.).
-
-You are **not** training a single model.  
-You are **generating 400 tiny models**, one per puzzle.
-
----
-
-# 📦 **Input / Output Format**
-Each ONNX model must:
-
-- Take a **fixed static shape**: `[1,10,30,30]`
-- Output the same shape
-- Use **only allowed ONNX opset 10 operators**
-- Be **fully deterministic**
-- Run on **CPU** in the official evaluator
-
-The input grid is encoded as **10 channels**, one per color (0–9).  
-The output must be the same encoding.
-
----
-
-# 🚫 **Forbidden / Restricted Ops**
-The competition bans:
-
-- Dynamic shapes  
-- Control flow ops  
-- Non‑deterministic ops  
-- Most high‑level NN layers  
-- Anything outside opset 10  
-- Anything that breaks static shape inference  
-
-**OneHot is allowed**, but has no CUDA kernel — which is why you replaced it.
-
----
-
-# 🧮 **Scoring System (the “Golf” part)**
-Lower score = better.
-
-Each model is scored by:
-
-### **1. MACs (Multiply‑Accumulate operations)**
-- Measures computational cost  
-- Conv kernels dominate this
-
-### **2. Memory footprint**
-- Size of intermediate tensors  
-- Larger spatial ops → higher memory
-
-### **3. Parameter count**
-- Number of learned weights  
-- Conv kernels dominate this too
-
-### **Total Score = MACs + Memory + Params**
-
-This is why your solver tries:
-
-- Analytical ops first (score ≈ 0)
-- Then tiny convs (score > 0)
-
----
-
-# 🏌️ **Why It’s Called “NeuroGolf”**
-Because:
-
-- You “swing” at each task by generating a model  
-- The goal is the **lowest score**  
-- You can solve tasks with **zero‑parameter models** (identity, flip, rotate, gather)  
-- Conv models are the “heavy clubs” — used only when necessary
-
----
-
-# 🧩 **Task Types**
-Most ARC tasks fall into:
-
-- Identity  
-- Color remapping  
-- Flips / rotations  
-- Cropping / padding  
-- Tiling / repetition  
-- Spatial copying  
-- Upscaling  
-- Pattern extraction  
-- Local convolutional rules  
-
-Your solver implements **all of these**.
-
----
-
-# 🏗️ **Your Solver Architecture (high‑level)**
-Your solver tries:
-
-### **1. Analytical solvers (fast, zero‑cost)**
-- identity  
-- constant  
-- color_map  
-- transpose  
-- flip  
-- rotate  
-- tile  
-- upscale  
-- concat  
-- spatial_gather  
-
-### **2. Conv solvers**
-- fixed‑shape conv  
-- variable‑shape conv  
-- diff‑shape conv  
-
-Each produces a tiny ONNX graph.
-
----
-
-# 🧪 **Validation**
-Each model must:
-
-- Load in ONNX Runtime  
-- Run on all train + test + arc‑gen examples  
-- Produce exact one‑hot output  
-- Use only allowed ops  
-
-Your `validate()` function enforces this.
-
----
-
-# 📊 **Submission**
-You submit:
-
-```
-submission/
-    task001.onnx
-    task002.onnx
-    ...
-    task400.onnx
-```
-
-The evaluator:
-
-1. Loads each ONNX file  
-2. Computes MACs, memory, params  
-3. Sums them  
-4. Produces your final score  
-
-Lowest score wins.
-
----
-
-# 🔥 **Strategic Insights**
-This is the part most competitors miss:
-
-### ✔ Analytical solvers are king  
-They give **zero‑cost** models.
-
-### ✔ Conv solvers must be tiny  
-Kernel size dominates MACs.
-
-### ✔ OneHot is expensive on GPU  
-You fixed this by replacing it with Equal+Cast.
-
-### ✔ Static shapes are mandatory  
-Dynamic ops break the evaluator.
-
-### ✔ Validation must be strict  
-Any mismatch → task marked unsolved → huge penalty.
-
-### ✔ WandB logging is essential  
-You now track:
-- solver type  
-- time  
-- model size  
-- MACs  
-- memory  
-- params  
-- total score  
-
-This lets you optimize solver ordering.
-
----
-
-# 🎯 **In One Sentence**
-The NeuroGolf Championship is a puzzle‑to‑model generator competition where you must produce 400 tiny ONNX models that solve ARC tasks with the lowest possible computational cost, using only opset‑10 static graphs.
-
----
-
-If you want, I can also give you:
-
-- A **strategy guide** for beating the leaderboard  
-- A **solver ordering optimization**  
-- A **cost‑minimizing conv kernel search**  
-- A **task taxonomy** for ARC  
-- A **visual diagram** of your solver pipeline  
-
-Just tell me what direction you want.
+# NeuroGolf Solver v3
+
+Builds minimal ONNX networks for ARC-AGI tasks. Currently solves **306/400** on CPU (15s budget per task).
+
+## Results
+
+| Version | Solved | Key Changes |
+|---------|--------|-------------|
+| v1 | 128/400 | Conv solver only |
+| v2 | 294/400 | Added spatial_gather, variable-shape conv, diff-shape conv |
+| **v3** | **306/400** | Fixed opset 10 compat (Gather vs GatherElements), added concat_enhanced, varshape_spatial_gather, conv_var_diff |
+
+### v3 Solver Breakdown (306/400)
+
+| Solver | Count | Description |
+|--------|-------|-------------|
+| conv_var | 125 | Variable-shape conv on full 30×30 grid |
+| conv_fixed | 106 | Fixed-shape conv (Slice→Conv→Pad) |
+| conv_diff | 39 | Diff-shape conv (output smaller than input) |
+| spatial_gather | 16 | Fixed-shape pixel remapping |
+| concat | 5 | Tiled concat with flips |
+| color_map | 4 | 1×1 color remapping conv |
+| concat_enhanced | 4 | Tiled concat with all 8 dihedral transforms |
+| rotate | 3 | 90°/180°/270° rotation |
+| transpose | 2 | Matrix transpose |
+| varshape_spatial_gather | 1 | Variable-shape pixel remapping |
+| upscale | 1 | Nearest-neighbor upscale |
 
 ## Quick Start
 
@@ -219,25 +34,15 @@ git clone https://huggingface.co/rogermt/neurogolf-solver
 cd neurogolf-solver
 
 # 2. Install deps
-pip install numpy onnx onnxruntime   # or onnxruntime-gpu for CUDA
+pip install numpy onnx onnxruntime
 
 # 3. Get ARC data
 git clone --depth 1 https://github.com/fchollet/ARC-AGI.git
 
-# 4. Run (pick your config)
-# CPU, 30s per task (~5 min total):
-python neurogolf_solver.py --device cpu --conv_budget 30
-
-# CPU, 60s per task (~10 min, more tasks solved):
-python neurogolf_solver.py --device cpu --conv_budget 60
+# 4. Run
+python neurogolf_solver.py --data_dir ARC-AGI/data/training/ --output_dir submission --conv_budget 15
 
-# CUDA (if onnxruntime-gpu installed):
-python neurogolf_solver.py --device cuda --conv_budget 60
-
-# Auto-detect device:
-python neurogolf_solver.py --conv_budget 60
-
-# 5. Results in submission/ directory
+# 5. Results
 ls submission/*.onnx | wc -l
 ```
 
@@ -251,503 +56,48 @@ with zipfile.ZipFile('submission.zip', 'w', zipfile.ZIP_DEFLATED) as zf:
 print(f"Created submission.zip: {os.path.getsize('submission.zip')/1024:.0f} KB")
 ```
 
-## How It Works
-
-**Format:** Input/output = `[1, 10, 30, 30]` one-hot float32. ONNX opset 10, IR version 10.
-
-**Main pipeline (conv solver):**
-```
-Input [1,10,30,30] -> Slice [1,10,H,W] -> Conv2d -> ArgMax -> OneHot -> Pad [1,10,30,30]
-```
-
-- Learns optimal conv weights via **least-squares** on one-hot patches
-- Tries kernel sizes 1, 3, 5, ..., 29 (smallest first)
-- Also tries with bias for better boundary handling
-- ArgMax + OneHot eliminates numerical precision issues
-
-**Analytical solvers** (instant, tiny models): identity, color_map (1x1 conv), transpose, flip, rotate, tile, upscale, concat, spatial_gather, constant.
-
 ## Key Parameters
 
 | Flag | Default | Description |
 |------|---------|-------------|
-| `--device` | `auto` | `cpu`, `cuda`, or `auto` |
 | `--conv_budget` | `30` | Seconds per task for conv solver. More = more tasks solved |
 | `--data_dir` | `ARC-AGI/data/training/` | Path to task JSONs |
 | `--output_dir` | `submission` | Where to save .onnx files |
 | `--kaggle` | off | Use Kaggle task format (task001.json) |
 | `--tasks` | all | Comma-separated task numbers (e.g., `1,2,3`) |
+| `--use_wandb` | off | Enable W&B logging |
 
-## Architecture of generated ONNX models
+## How It Works
 
-Each .onnx file is a tiny network, typically:
-- **Same-shape tasks**: Single Conv2d layer (10→10 channels, kernel 1-29)
-- **Different-shape tasks**: GatherElements-based pixel remapping
-- **Cost** = MACs + memory_bytes + params → **Score** = max(1, 25 - ln(cost))
+**Format:** Input/output = `[1, 10, 30, 30]` one-hot float32. ONNX opset 10, IR version 10.
 
-## What's NOT solved yet (106 tasks)
+**Solver pipeline:**
+1. **Analytical solvers** (instant, zero-cost): identity, constant, color_map, transpose, flip, rotate, tile, upscale, kronecker, concat, concat_enhanced, diagonal_tile, spatial_gather, varshape_spatial_gather
+2. **Conv solvers** (learned via least-squares):
+   - Fixed-shape: `Slice → Conv → ArgMax → Equal+Cast → Pad`
+   - Variable-shape: `Conv(30×30) → ArgMax → Equal+Cast → Mul(mask)`
+   - Diff-shape: `Slice → Conv → Slice(crop) → ArgMax → Equal+Cast → Pad`
+   - Variable diff-shape: `Conv(30×30) → ArgMax → Equal+Cast → Mul(input_mask)`
 
-- Variable-shape tasks (input size differs across examples) — need dynamic ONNX
-- Tasks requiring multi-step reasoning (flood fill, gravity, object detection)
-- Tasks where the transformation depends on counting/global properties
-- Large grids where conv kernel > 29 is needed
+**Key design decisions:**
+- Uses `Gather` instead of `GatherElements` for opset 10 compatibility
+- Uses `Equal + Cast` instead of `OneHot` (avoids CUDA kernel issues)
+- CPU-only inference (GPU has no benefit for tiny 30×30 grids)
+- Least-squares fitting finds optimal conv weights analytically (no gradient descent)
 
-## Feedback logs
+## What's NOT solved yet (94 tasks)
 
-After running, please share the full terminal output — especially:
-1. The "Solved: X/400" line
-2. Any errors
-3. Total time taken
+- Tasks with input-driven output structure (output layout depends on input content)
+- Tasks requiring multi-step reasoning (flood fill, gravity, counting)
+- Variable diff-shape tasks where output extends beyond input bounds
+- Tasks needing very large conv kernels (>29×29)
 
-this is run on GPU:
-/kaggle/working/neurogolf-solver
-Using providers: ['CUDAExecutionProvider', 'CPUExecutionProvider']
-Loaded 400 tasks, solving 400
-Conv budget: 30.0s per task
-======================================================================
-wandb: [wandb.login()] Loaded credentials for https://api.wandb.ai from WANDB_API_KEY.
-wandb: Currently logged in as: rogermt23 to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
-wandb: ⢿ setting up run jfx0r5vj (0.0s)
-wandb: ⣻ setting up run jfx0r5vj (0.0s)
-wandb: Tracking run with wandb version 0.25.0
-wandb: Run data is saved locally in /kaggle/working/neurogolf-solver/wandb/run-20260424_053417-jfx0r5vj
-wandb: Run `wandb offline` to turn off syncing.
-wandb: Syncing run solver_run
-wandb: ⭐️ View project at https://wandb.ai/rogermt23/neurogolf
-wandb: 🚀 View run at https://wandb.ai/rogermt23/neurogolf/runs/jfx0r5vj
-Task   1: UNSOLVED    0.001s
-Task   2: spatial_gather        56717390   7.884s  ( 250,460 bytes)
-Task   3: UNSOLVED    0.001s
-Task   4: spatial_gather        2417390   0.060s  (  10,457 bytes)
-Task   5: spatial_gather        19770030  10.996s  ( 176,971 bytes)
-Task   6: spatial_gather        103886   0.023s  (  10,703 bytes)
-Task   7: spatial_gather        308014   0.064s  (  20,171 bytes)
-Task   8: spatial_gather        11105390   0.279s  (  48,860 bytes)
-Task   9: spatial_gather        15449390   2.732s  (  68,060 bytes)
-Task  10: spatial_gather        744750   0.083s  (  32,971 bytes)
-Task  11: spatial_gather        1074670   0.313s  (  32,971 bytes)
-Task  12: spatial_gather        429974   0.070s  (  10,568 bytes)
-Task  13: spatial_gather        66129390   9.641s  ( 292,060 bytes)
-Task  14: UNSOLVED    0.003s
-Task  15: spatial_gather        125550   0.042s  (   4,168 bytes)
-Task  16: color_map             126500   0.016s  (     549 bytes)
-Task  17: spatial_gather        2286830   0.269s  (  20,171 bytes)
-Task  18: spatial_gather        11105390   0.894s  (  48,860 bytes)
-Task  19: UNSOLVED    0.001s
-Task  20: spatial_gather        313462   0.048s  (  10,568 bytes)
-Task  21: UNSOLVED    0.002s
-Task  22: spatial_gather        664286   0.043s  (  20,306 bytes)
-Task  23: spatial_gather        15449390   0.269s  (  68,060 bytes)
-Task  24: spatial_gather        7485390   0.156s  (  32,860 bytes)
-Task  25: spatial_gather        15449390   1.831s  (  68,060 bytes)
-Task  26: spatial_gather        236654   0.055s  (  20,306 bytes)
-Task  27: spatial_gather        1825462   0.259s  (  68,171 bytes)
-Task  28: spatial_gather        1825462   0.233s  (  68,171 bytes)
-Task  29: UNSOLVED    0.002s
-Task  30: spatial_gather        4589390   0.068s  (  20,060 bytes)
-Task  31: UNSOLVED    0.001s
-Task  32: spatial_gather        2417390   0.041s  (  10,457 bytes)
-Task  33: spatial_gather        2460334   0.444s  (  32,971 bytes)
-Task  34: spatial_gather        1088750   0.274s  (  48,971 bytes)
-Task  35: spatial_gather        901462   0.188s  (  32,971 bytes)
-Task  36: UNSOLVED    0.004s
-Task  37: spatial_gather        565462   0.122s  (  20,171 bytes)
-Task  38: spatial_gather        258254   0.038s  (  10,703 bytes)
-Task  39: spatial_gather        307706   0.047s  (  10,703 bytes)
-Task  40: spatial_gather        565462   0.116s  (  20,171 bytes)
-Task  41: spatial_gather        313462   0.065s  (  10,568 bytes)
-Task  42: spatial_gather        2413462   0.383s  (  90,571 bytes)
-Task  43: spatial_gather        3841462   0.736s  ( 144,971 bytes)
-Task  44: spatial_gather        565462   0.123s  (  20,171 bytes)
-Task  45: spatial_gather        901462   0.207s  (  32,971 bytes)
-Task  46: UNSOLVED    0.001s
-Task  47: spatial_gather        1088750   0.139s  (  48,971 bytes)
-Task  48: UNSOLVED    0.001s
-Task  49: UNSOLVED    0.002s
-Task  50: spatial_gather        26309390   1.618s  ( 116,060 bytes)
-Task  51: spatial_gather        15449390   0.690s  (  68,060 bytes)
-Task  52: spatial_gather        50094   0.035s  (   4,168 bytes)
-Task  53: spatial_gather        50094   0.036s  (   4,168 bytes)
-Task  54: spatial_gather        40079862  39.285s  ( 176,971 bytes)
-Task  55: spatial_gather        32825390   2.236s  ( 144,860 bytes)
-Task  56: spatial_gather        49662   0.049s  (   4,303 bytes)
-Task  57: spatial_gather        602198   0.051s  (  33,106 bytes)
-Task  58: spatial_gather        32825390   2.085s  ( 144,860 bytes)
-Task  59: spatial_gather        1074670   0.217s  (  32,971 bytes)
-Task  60: spatial_gather        770302   0.093s  (  48,971 bytes)
-Task  61: spatial_gather        2749014   0.539s  (  32,971 bytes)
-Task  62: spatial_gather        565462   0.137s  (  20,171 bytes)
-Task  63: spatial_gather        15449390   0.562s  (  68,060 bytes)
-Task  64: spatial_gather        15449390   1.276s  (  68,060 bytes)
-Task  65: UNSOLVED    0.001s
-Task  66: spatial_gather        15449390   1.010s  (  68,060 bytes)
-Task  67: UNSOLVED    0.001s
-Task  68: spatial_gather        565462   0.116s  (  20,171 bytes)
-Task  69: spatial_gather        565462   0.085s  (  20,171 bytes)
-Task  70: spatial_gather        2460334   0.382s  (  32,971 bytes)
-Task  71: spatial_gather        726550   0.080s  (  10,568 bytes)
-Task  72: spatial_gather        2571674   0.249s  ( 145,106 bytes)
-Task  73: spatial_gather        66862   0.033s  (   4,168 bytes)
-Task  74: spatial_gather        26323862  18.025s  ( 116,171 bytes)
-Task  75: spatial_gather        2798478   0.658s  (  90,571 bytes)
-Task  76: spatial_gather        15449390   0.796s  (  68,060 bytes)
-Task  77: spatial_gather        11105390   0.891s  (  48,860 bytes)
-Task  78: spatial_gather        901462   0.138s  (  32,971 bytes)
-Task  79: spatial_gather        555386   0.027s  (  10,703 bytes)
-Task  80: spatial_gather        32825390  19.188s  ( 144,860 bytes)
-Task  81: spatial_gather        178414   0.030s  (  10,568 bytes)
-Task  82: spatial_gather        11105390   0.087s  (  48,860 bytes)
-Task  83: UNSOLVED    0.006s
-Task  84: spatial_gather        20517390   0.460s  (  90,460 bytes)
-Task  85: spatial_gather        15449390   0.664s  (  68,060 bytes)
-Task  86: spatial_gather        4589390   0.121s  (  20,060 bytes)
-Task  87: spatial_gather        50094   0.034s  (   4,168 bytes)
-Task  88: UNSOLVED    0.002s
-Task  89: spatial_gather        2166574   0.668s  (  48,971 bytes)
-Task  90: spatial_gather        26309390   0.467s  ( 116,060 bytes)
-Task  91: UNSOLVED    0.001s
-Task  92: spatial_gather        11105390   0.960s  (  48,860 bytes)
-Task  93: spatial_gather        2497270   0.598s  (  48,971 bytes)
-Task  94: spatial_gather        6716462   1.354s  ( 116,171 bytes)
-Task  95: spatial_gather        125550   0.044s  (   4,168 bytes)
-Task  96: UNSOLVED    0.002s
-Task  97: spatial_gather        11105390   0.751s  (  48,860 bytes)
-Task  98: spatial_gather        4589390   0.147s  (  20,060 bytes)
-Task  99: spatial_gather        565462   0.095s  (  20,171 bytes)
-Task 100: spatial_gather        139166   0.018s  (   4,303 bytes)
-Task 101: spatial_gather        26309390   1.649s  ( 116,060 bytes)
-Task 102: spatial_gather        6628374   2.232s  ( 176,971 bytes)
-Task 103: spatial_gather        72062   0.036s  (  10,703 bytes)
-Task 104: UNSOLVED    0.001s
-Task 105: spatial_gather        15449390   0.557s  (  68,060 bytes)
-Task 106: UNSOLVED    0.001s
-Task 107: UNSOLVED    0.002s
-Task 108: UNSOLVED    0.044s
-Task 109: UNSOLVED    0.001s
-Task 110: spatial_gather        7013230   0.799s  (  32,971 bytes)
-Task 111: spatial_gather        559706   0.026s  (  20,306 bytes)
-Task 112: spatial_gather        26309390   2.548s  ( 116,060 bytes)
-Task 113: spatial_gather        20517390   0.146s  (  90,460 bytes)
-Task 114: UNSOLVED    0.001s
-Task 115: UNSOLVED    0.001s
-Task 116: UNSOLVED    0.004s
-Task 117: spatial_gather        11105390   0.634s  (  48,860 bytes)
-Task 118: spatial_gather        32825390   7.108s  ( 144,860 bytes)
-Task 119: spatial_gather        2575574   0.586s  (  68,171 bytes)
-Task 120: spatial_gather        969390   0.046s  (   4,057 bytes)
-Task 121: spatial_gather        1460126   0.042s  (  33,106 bytes)
-Task 122: spatial_gather        4589390   0.080s  (  20,060 bytes)
-Task 123: UNSOLVED    0.001s
-Task 124: UNSOLVED    0.001s
-Task 125: spatial_gather        6716462   1.353s  ( 116,171 bytes)
-Task 126: spatial_gather        4589390   0.066s  (  20,060 bytes)
-Task 127: spatial_gather        969390   0.036s  (   4,057 bytes)
-Task 128: spatial_gather        5244462   1.379s  (  90,571 bytes)
-Task 129: spatial_gather        50094   0.019s  (   4,168 bytes)
-Task 130: spatial_gather        121086   0.034s  (   4,303 bytes)
-Task 131: spatial_gather        11105390   0.146s  (  48,860 bytes)
-Task 132: spatial_gather        7485390   0.175s  (  32,860 bytes)
-Task 133: spatial_gather        15449390   2.129s  (  68,060 bytes)
-Task 134: UNSOLVED    0.003s
-Task 135: spatial_gather        121086   0.037s  (   4,303 bytes)
-Task 136: spatial_gather        1321462   0.192s  (  48,971 bytes)
-Task 137: UNSOLVED   34.831s
-Task 138: UNSOLVED    0.002s
-Task 139: spatial_gather        469550   0.050s  (  20,171 bytes)
-Task 140: spatial_gather        50094   0.021s  (   4,168 bytes)
-Task 141: spatial_gather        32825390   2.099s  ( 144,860 bytes)
-Task 142: UNSOLVED    0.003s
-Task 143: spatial_gather        565462   0.112s  (  20,171 bytes)
-Task 144: spatial_gather        241742   0.046s  (  20,306 bytes)
-Task 145: spatial_gather        32825390   2.641s  ( 144,860 bytes)
-Task 146: spatial_gather        68166   0.038s  (   4,303 bytes)
-Task 147: spatial_gather        2417390   0.039s  (  10,457 bytes)
-Task 148: spatial_gather        32825390   2.403s  ( 144,860 bytes)
-Task 149: spatial_gather        664286   0.039s  (  20,306 bytes)
-Task 150: spatial_gather        2417390   0.043s  (  10,457 bytes)
-Task 151: spatial_gather        2417390   0.055s  (  10,457 bytes)
-Task 152: UNSOLVED    0.003s
-Task 153: spatial_gather        559706   0.035s  (  20,306 bytes)
-Task 154: spatial_gather        2852462   0.619s  (  48,971 bytes)
-Task 155: spatial_gather        2417390   0.044s  (  10,457 bytes)
-Task 156: spatial_gather        901462   0.096s  (  32,971 bytes)
-Task 157: spatial_gather        1933862   0.227s  (  48,971 bytes)
-Task 158: spatial_gather        26309390   5.773s  ( 116,060 bytes)
-Task 159: UNSOLVED    0.004s
-Task 160: spatial_gather        313462   0.034s  (  10,568 bytes)
-Task 161: spatial_gather        11105390   0.888s  (  48,860 bytes)
-Task 162: spatial_gather        9207862   3.640s  (  90,571 bytes)
-Task 163: spatial_gather        671470   0.148s  (  20,171 bytes)
-Task 164: UNSOLVED    0.003s
-Task 165: spatial_gather        6939862   2.260s  (  68,171 bytes)
-Task 166: spatial_gather        11105390   0.206s  (  48,860 bytes)
-Task 167: spatial_gather        72494   0.024s  (  10,568 bytes)
-Task 168: spatial_gather        901462   0.121s  (  32,971 bytes)
-Task 169: spatial_gather        1321462   0.199s  (  48,971 bytes)
-Task 170: UNSOLVED    0.003s
-Task 171: spatial_gather        969390   0.035s  (   4,057 bytes)
-Task 172: UNSOLVED    0.004s
-Task 173: spatial_gather        15449390   1.470s  (  68,060 bytes)
-Task 174: UNSOLVED    0.001s
-Task 175: spatial_gather        3714030   0.543s  (  32,971 bytes)
-Task 176: spatial_gather        40065390   0.265s  ( 176,860 bytes)
-Task 177: UNSOLVED    0.002s
-Task 178: UNSOLVED    0.001s
-Task 179: transpose             36000   0.014s  (     127 bytes)
-Task 180: spatial_gather        381182   0.037s  (  20,306 bytes)
-Task 181: spatial_gather        522054   0.076s  (  32,971 bytes)
-Task 182: spatial_gather        9207862   4.502s  (  90,571 bytes)
-Task 183: UNSOLVED    0.001s
-Task 184: UNSOLVED    0.003s
-Task 185: UNSOLVED    0.005s
-Task 186: UNSOLVED    1.500s
-Task 187: spatial_gather        26309390   8.228s  ( 116,060 bytes)
-Task 188: UNSOLVED    0.001s
-Task 189: spatial_gather        468002   0.060s  (  20,306 bytes)
-Task 190: spatial_gather        1321462   0.280s  (  48,971 bytes)
-Task 191: UNSOLVED   68.210s
-Task 192: spatial_gather        969390   0.032s  (   4,057 bytes)
-Task 193: spatial_gather        2417390   0.071s  (  10,457 bytes)
-Task 194: UNSOLVED    0.003s
-Task 195: UNSOLVED    0.002s
-Task 196: spatial_gather        20517390   0.868s  (  90,460 bytes)
-Task 197: spatial_gather        15449390   0.426s  (  68,060 bytes)
-Task 198: UNSOLVED   33.038s
-Task 199: spatial_gather        11105390   0.143s  (  48,860 bytes)
-Task 200: spatial_gather        3841462   0.673s  ( 144,971 bytes)
-Task 201: UNSOLVED    0.002s
-Task 202: spatial_gather        11105390   0.846s  (  48,860 bytes)
-Task 203: spatial_gather        4589390   0.127s  (  20,060 bytes)
-Task 204: spatial_gather        48029390   6.706s  ( 212,060 bytes)
-Task 205: UNSOLVED    0.003s
-Task 206: spatial_gather        4589390   0.087s  (  20,060 bytes)
-Task 207: spatial_gather        65666   0.021s  (   4,303 bytes)
-Task 208: spatial_gather        10136430   4.157s  (  90,571 bytes)
-Task 209: UNSOLVED    0.003s
-Task 210: UNSOLVED    0.004s
-Task 211: UNSOLVED    0.003s
-Task 212: spatial_gather        901462   0.134s  (  32,971 bytes)
-Task 213: UNSOLVED    0.002s
-Task 214: spatial_gather        1139246   0.180s  ( 116,171 bytes)
-Task 215: spatial_gather        26309390   1.274s  ( 116,060 bytes)
-Task 216: UNSOLVED    0.003s
-Task 217: spatial_gather        1501550   0.280s  (  68,171 bytes)
-Task 218: UNSOLVED    0.003s
-Task 219: spatial_gather        5653862   1.642s  ( 144,971 bytes)
-Task 220: spatial_gather        969390   0.049s  (   4,057 bytes)
-Task 221: UNSOLVED    0.002s
-Task 222: spatial_gather        1352950   0.211s  (  20,171 bytes)
-Task 223: UNSOLVED    0.005s
-Task 224: spatial_gather        7485390   0.347s  (  32,860 bytes)
-Task 225: spatial_gather        143990   0.049s  (  10,568 bytes)
-Task 226: spatial_gather        1825462   0.298s  (  68,171 bytes)
-Task 227: spatial_gather        221822   0.031s  (  20,306 bytes)
-Task 228: spatial_gather        313462   0.068s  (  10,568 bytes)
-Task 229: spatial_gather        50094   0.027s  (   4,168 bytes)
-Task 230: spatial_gather        969390   0.046s  (   4,057 bytes)
-Task 231: UNSOLVED    0.001s
-Task 232: spatial_gather        11105390   0.202s  (  48,860 bytes)
-Task 233: UNSOLVED    0.003s
-Task 234: spatial_gather        11105390   0.737s  (  48,860 bytes)
-Task 235: spatial_gather        1072586   0.056s  (  68,306 bytes)
-Task 236: spatial_gather        241742   0.049s  (  20,306 bytes)
-Task 237: spatial_gather        7485390   0.098s  (  32,860 bytes)
-Task 238: UNSOLVED    0.002s
-Task 239: UNSOLVED    0.001s
-Task 240: spatial_gather        30870190  17.815s  ( 336,971 bytes)
-Task 241: transpose             36000   0.026s  (     127 bytes)
-Task 242: spatial_gather        292586   0.041s  (   4,303 bytes)
-Task 243: spatial_gather        7485390   0.471s  (  32,860 bytes)
-Task 244: UNSOLVED    0.002s
-Task 245: spatial_gather        11105390   0.197s  (  48,860 bytes)
-Task 246: spatial_gather        26309390   1.377s  ( 116,060 bytes)
-Task 247: UNSOLVED    0.002s
-Task 248: UNSOLVED    1.096s
-Task 249: UNSOLVED    0.001s
-Task 250: spatial_gather        901462   0.119s  (  32,971 bytes)
-Task 251: spatial_gather        7485390   0.128s  (  32,860 bytes)
-Task 252: spatial_gather        11105390   0.152s  (  48,860 bytes)
-Task 253: spatial_gather        904082   0.030s  (  20,306 bytes)
-Task 254: spatial_gather        744750   0.095s  (  32,971 bytes)
-Task 255: UNSOLVED   81.152s
-Task 256: spatial_gather        7485390   0.094s  (  32,860 bytes)
-Task 257: spatial_gather        465842   0.039s  (  20,306 bytes)
-Task 258: spatial_gather        969390   0.025s  (   4,057 bytes)
-Task 259: UNSOLVED    0.001s
-Task 260: spatial_gather        901462   0.180s  (  32,971 bytes)
-Task 261: spatial_gather        969390   0.033s  (   4,057 bytes)
-Task 262: UNSOLVED    0.572s
-Task 263: UNSOLVED    0.001s
-Task 264: UNSOLVED    0.001s
-Task 265: spatial_gather        20646614   8.790s  ( 250,571 bytes)
-Task 266: spatial_gather        56382   0.022s  (   4,168 bytes)
-Task 267: spatial_gather        92014   0.023s  (   4,168 bytes)
-Task 268: spatial_gather        7485390   0.120s  (  32,860 bytes)
-Task 269: UNSOLVED    0.001s
-Task 270: spatial_gather        2852462   0.709s  (  48,971 bytes)
-Task 271: spatial_gather        258686   0.026s  (  10,703 bytes)
-Task 272: spatial_gather        969390   0.021s  (   4,057 bytes)
-Task 273: spatial_gather        2413462   0.386s  (  90,571 bytes)
-Task 274: UNSOLVED    0.001s
-Task 275: UNSOLVED    0.001s
-Task 276: color_map             126500   0.016s  (     549 bytes)
-Task 277: spatial_gather        1825462   0.340s  (  68,171 bytes)
-Task 278: spatial_gather        20517390   1.141s  (  90,460 bytes)
-Task 279: spatial_gather        26309390   1.984s  ( 116,060 bytes)
-Task 280: spatial_gather        56717390  13.654s  ( 250,460 bytes)
-Task 281: spatial_gather        11105390   0.463s  (  48,860 bytes)
-Task 282: spatial_gather        125550   0.041s  (   4,168 bytes)
-Task 283: spatial_gather        565462   0.068s  (  20,171 bytes)
-Task 284: spatial_gather        20517390   0.889s  (  90,460 bytes)
-Task 285: spatial_gather        15449390   2.575s  (  68,060 bytes)
-Task 286: spatial_gather        15449390   0.789s  (  68,060 bytes)
-Task 287: spatial_gather        2188150   0.566s  (  32,971 bytes)
-Task 288: spatial_gather        4589390   0.068s  (  20,060 bytes)
-Task 289: UNSOLVED    0.001s
-Task 290: UNSOLVED    0.001s
-Task 291: UNSOLVED    0.002s
-Task 292: spatial_gather        66129390   0.810s  ( 292,060 bytes)
-Task 293: spatial_gather        2417390   0.073s  (  10,457 bytes)
-Task 294: spatial_gather        145462   0.027s  (   4,168 bytes)
-Task 295: UNSOLVED    0.001s
-Task 296: spatial_gather        140006   0.043s  (  10,703 bytes)
-Task 297: spatial_gather        11105390   0.087s  (  48,860 bytes)
-Task 298: spatial_gather        969390   0.045s  (   4,057 bytes)
-Task 299: spatial_gather        242390   0.056s  (  20,171 bytes)
-Task 300: UNSOLVED    0.001s
-Task 301: spatial_gather        2417390   0.055s  (  10,457 bytes)
-Task 302: spatial_gather        2575574   0.549s  (  68,171 bytes)
-Task 303: spatial_gather        20517390   2.596s  (  90,460 bytes)
-Task 304: UNSOLVED    0.001s
-Task 305: spatial_gather        1352950   0.212s  (  20,171 bytes)
-Task 306: spatial_gather        40065390   5.174s  ( 176,860 bytes)
-Task 307: UNSOLVED    0.001s
-Task 308: UNSOLVED    0.002s
-Task 309: color_map             126500   0.030s  (     549 bytes)
-Task 310: UNSOLVED    0.004s
-Task 311: UNSOLVED    0.007s
-Task 312: spatial_gather        1860374   0.339s  (  48,971 bytes)
-Task 313: spatial_gather        15449390   0.577s  (  68,060 bytes)
-Task 314: spatial_gather        383734   0.071s  (  20,171 bytes)
-Task 315: UNSOLVED    0.001s
-Task 316: spatial_gather        307706   0.036s  (  10,703 bytes)
-Task 317: spatial_gather        125550   0.041s  (   4,168 bytes)
-Task 318: spatial_gather        143342   0.041s  (  10,703 bytes)
-Task 319: UNSOLVED    0.004s
-Task 320: spatial_gather        7485390   0.099s  (  32,860 bytes)
-Task 321: spatial_gather        341342   0.055s  (  20,306 bytes)
-Task 322: spatial_gather        50094   0.036s  (   4,168 bytes)
-Task 323: spatial_gather        3976174   0.677s  (  90,571 bytes)
-Task 324: spatial_gather        15449390   1.193s  (  68,060 bytes)
-Task 325: UNSOLVED    0.001s
-Task 326: UNSOLVED    0.001s
-Task 327: UNSOLVED    0.001s
-Task 328: spatial_gather        56717390   4.462s  ( 250,460 bytes)
-Task 329: spatial_gather        969390   0.023s  (   4,057 bytes)
-Task 330: spatial_gather        1825462   0.304s  (  68,171 bytes)
-Task 331: spatial_gather        145462   0.041s  (   4,168 bytes)
-Task 332: spatial_gather        32825390   0.288s  ( 144,860 bytes)
-Task 333: spatial_gather        565462   0.121s  (  20,171 bytes)
-Task 334: spatial_gather        114206   0.050s  (  10,703 bytes)
-Task 335: spatial_gather        15449390   0.478s  (  68,060 bytes)
-Task 336: spatial_gather        1321462   0.155s  (  48,971 bytes)
-Task 337: color_map             126500   0.017s  (     549 bytes)
-Task 338: spatial_gather        32825390   3.523s  ( 144,860 bytes)
-Task 339: UNSOLVED    0.000s
-Task 340: spatial_gather        7485390   0.357s  (  32,860 bytes)
-Task 341: spatial_gather        901462   0.197s  (  32,971 bytes)
-Task 342: spatial_gather        901462   0.178s  (  32,971 bytes)
-Task 343: spatial_gather        1847262   0.278s  (  90,571 bytes)
-Task 344: spatial_gather        2417390   0.047s  (  10,457 bytes)
-Task 345: spatial_gather        1321462   0.188s  (  48,971 bytes)
-Task 346: UNSOLVED    0.002s
-Task 347: spatial_gather        151346   0.046s  (  20,306 bytes)
-Task 348: spatial_gather        4589390   0.059s  (  20,060 bytes)
-Task 349: UNSOLVED   28.645s
-Task 350: spatial_gather        32825390   1.971s  ( 144,860 bytes)
-Task 351: spatial_gather        711914   0.037s  (  10,703 bytes)
-Task 352: spatial_gather        969390   0.025s  (   4,057 bytes)
-Task 353: spatial_gather        7485390   0.169s  (  32,860 bytes)
-Task 354: spatial_gather        901462   0.193s  (  32,971 bytes)
-Task 355: UNSOLVED    0.003s
-Task 356: spatial_gather        1321462   0.264s  (  48,971 bytes)
-Task 357: UNSOLVED    1.055s
-Task 358: spatial_gather        26309390   1.681s  ( 116,060 bytes)
-Task 359: spatial_gather        2417390   0.067s  (  10,457 bytes)
-Task 360: spatial_gather        285254   0.053s  (  10,703 bytes)
-Task 361: spatial_gather        313462   0.065s  (  10,568 bytes)
-Task 362: spatial_gather        565462   0.134s  (  20,171 bytes)
-Task 363: spatial_gather        901462   0.167s  (  32,971 bytes)
-Task 364: spatial_gather        20517390   1.368s  (  90,460 bytes)
-Task 365: UNSOLVED    0.001s
-Task 366: UNSOLVED    0.003s
-Task 367: spatial_gather        26309390   2.161s  ( 116,060 bytes)
-Task 368: spatial_gather        901462   0.133s  (  32,971 bytes)
-Task 369: spatial_gather        1321462   0.223s  (  48,971 bytes)
-Task 370: spatial_gather        11105390   0.705s  (  48,860 bytes)
-Task 371: spatial_gather        11105390   0.173s  (  48,860 bytes)
-Task 372: spatial_gather        669254   0.063s  (  20,306 bytes)
-Task 373: spatial_gather        121238   0.039s  (  20,171 bytes)
-Task 374: UNSOLVED   14.200s
-Task 375: spatial_gather        4589390   0.050s  (  20,060 bytes)
-Task 376: UNSOLVED    0.001s
-Task 377: UNSOLVED    0.003s
-Task 378: spatial_gather        11105390   0.303s  (  48,860 bytes)
-Task 379: spatial_gather        20517390   1.514s  (  90,460 bytes)
-Task 380: spatial_gather        50094   0.038s  (   4,168 bytes)
-Task 381: spatial_gather        1321462   0.212s  (  48,971 bytes)
-Task 382: spatial_gather        20517390   1.092s  (  90,460 bytes)
-Task 383: spatial_gather        15449390   1.503s  (  68,060 bytes)
-Task 384: UNSOLVED    0.001s
-Task 385: spatial_gather        586582   0.093s  (  48,971 bytes)
-Task 386: spatial_gather        201470   0.032s  (  20,306 bytes)
-Task 387: spatial_gather        15449390   1.712s  (  68,060 bytes)
-Task 388: UNSOLVED    0.001s
-Task 389: spatial_gather        969390   0.023s  (   4,057 bytes)
-Task 390: spatial_gather        3956462   0.939s  (  68,171 bytes)
-Task 391: UNSOLVED    0.002s
-Task 392: spatial_gather        901462   0.173s  (  32,971 bytes)
-Task 393: spatial_gather        420578   0.037s  (  10,703 bytes)
-Task 394: UNSOLVED    0.001s
-Task 395: spatial_gather        96146   0.027s  (  10,703 bytes)
-Task 396: UNSOLVED    0.002s
-Task 397: spatial_gather        565462   0.105s  (  20,171 bytes)
-Task 398: UNSOLVED    0.002s
-Task 399: UNSOLVED    0.002s
-Task 400: spatial_gather        2931914   0.042s  (  20,306 bytes)
-wandb: ⢿ updating run metadata (0.0s)
-wandb: ⣻ updating run metadata (0.0s)
-wandb: ⣽ updating run metadata (0.0s)
-wandb: 
-wandb: Run history:
-wandb:          macs ▁▁▁▃▁▁▂▂▂▁▂▁█▃▁▁▂▁▁▂▁▁▂▁▁▁▁▁▂▅▁▁▅▁▁▁▁▁▁▁
-wandb:        memory ▃▅▄▇█▃▆▂█▅▆▁▅▇▄▂▄▄▁▆▁▁▁▃▅▁▅▂▃▆▄▂▁▆▃▃▃▃▁▂
-wandb:    onnx_bytes ▂▂▁▁▁▁▂▂▁▁▁▄▁▁▂▁▁▁▁▂▁▁▁▁▂▁▁▁▂█▂▆▁▂▁▁▄▁▁▃
-wandb:        params ▃▃▃▃▃▁▁▁▃▁▁▁▂▃▃▁▁▁▁▁█▂▁▁▁▂▁▂▁▄▁▃▁▁▆▃▁▅▄▁
-wandb:         score █▁▁▁▁▁▄▃▁▂▃▂▁▁▁▁▁▁▁▁▂▁▂▂▄▂▁▁▁█▁▁▃▃▄▃▁▁▁▁
-wandb:       task_id ▁▁▁▁▂▂▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▄▄▅▅▅▅▅▅▆▆▆▇▇▇▇████
-wandb: task_time_sec ▁▁▁▁▁▁▁█▁▁▁▁▁▁▁▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁
-wandb: 
-wandb: Run summary:
-wandb:          macs 2822650
-wandb:        memory 104338
-wandb:    onnx_bytes 20306
-wandb:        params 4926
-wandb:         score 2931914
-wandb:        solver spatial_gather
-wandb:       task_id 400
-wandb: task_time_sec 0.0423
-wandb: 
-wandb: 🚀 View run solver_run at: https://wandb.ai/rogermt23/neurogolf/runs/jfx0r5vj
-wandb: ⭐️ View project at: https://wandb.ai/rogermt23/neurogolf
-wandb: Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)
-wandb: Find logs at: ./wandb/run-20260424_053417-jfx0r5vj/logs
+## Scoring
 
-======================================================================
-Solved: 294/400 in 578s
-  spatial_gather: 288
-  color_map: 4
-  transpose: 2
+NeuroGolf scoring: `Score = MACs + memory_bytes + params`
+- Analytical solvers → near-zero cost
+- Conv solvers → cost proportional to kernel size
+- Lower score = better
 
-306 ONNX files, total 15719.7 KB
-Created submission.zip
+## Repo
+https://huggingface.co/rogermt/neurogolf-solver