Upload README.md

README.md

@@ -21,7 +21,7 @@ This repository contains the results of comparing different bidding algorithms f
 | **Threshold** | Static | Fixed bid if pCTR > threshold, else 0 |
 | **MPC** | Adaptive | Model Predictive Control maximizing expected value |
 
-## Results
+## Results on Synthetic Data
 
 | Algorithm | Clicks | CTR | Budget Used | CPC | Efficiency |
 |-----------|--------|-----|-------------|-----|------------|
@@ -31,13 +31,28 @@ This repository contains the results of comparing different bidding algorithms f
 | **Threshold** | 110 | 0.5500 | 100.00% | 90.91 | 110.00 |
 | **ORTB** | 85 | 0.5152 | 99.87% | 117.50 | 85.11 |
 
+## Results on Real Criteo Data (100K rows, CTR=25.7%)
+
+| Algorithm | Clicks | CTR | Budget Used | CPC |
+|-----------|--------|-----|-------------|-----|
+| **DualOGD** | **584** | 0.2421 | 100.00% | **8.56** |
+| **Linear** | 63 | 0.2593 | 100.00% | 79.36 |
+| **ORTB** | 47 | 0.2597 | 99.94% | 106.32 |
+| **Threshold** | 41 | 0.2470 | 99.60% | 121.46 |
+
 ## Key Findings
 
-1. **DualOGD dominates** with 1331 clicks — 8× better than Linear, 15× better than ORTB
-2. **Adaptive algorithms (DualOGD, MPC)** significantly outperform static approaches
+1. **DualOGD dominates on both datasets** — 9× better than Linear on real Criteo data
+2. **Adaptive algorithms significantly outperform static approaches** in first-price auctions
 3. **ORTB performs poorly** in first-price auctions (designed for second-price)
-4. **Threshold has highest CTR (0.55)** but low volume — very selective
-5. All algorithms achieved ~100% budget utilization
+4. **All algorithms achieve ~100% budget utilization** as required
+5. **DualOGD has the lowest CPC** — most cost-efficient
+
+## Files
+
+- `train.py` — Full training and comparison script
+- `results.json` — Synthetic data results
+- `results_real.json` — Real Criteo data results
 
 ## References