File size: 8,441 Bytes
2e0d61d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
"""
First-Price Auction Simulator for RTB Bidding Benchmark

Simulates a stream of first-price auctions with:
  - Impression features (from Criteo_x4 for CTR)
  - Click labels (ground truth from Criteo)
  - Synthetic market prices (competing bids) conditioned on features
  - Full information feedback (observes all competing bids)

Used to evaluate bidding algorithms under controlled conditions.
"""
import numpy as np
from collections import defaultdict


class FirstPriceAuctionSimulator:
    """
    Simulates repeated first-price auctions.
    
    In a first-price auction:
      - Each bidder submits a sealed bid
      - Highest bidder wins and pays their bid
      - Losing bidders pay nothing
    
    We simulate the "advertiser" (our bidder) vs. a pool of competing bidders.
    The maximum competing bid d_t follows a distribution conditioned on features.
    """
    
    def __init__(
        self,
        features,
        pctr_true,
        click_labels,
        value_per_click=50.0,
        market_price_config=None,
        seed=42
    ):
        """
        Args:
            features: (N, D) impression feature matrix
            pctr_true: (N,) true or predicted CTR values
            click_labels: (N,) binary click labels
            value_per_click: Value of each click in currency
            market_price_config: Dict controlling price generation
            seed: Random seed
        """
        self.features = features
        self.pctr_true = pctr_true
        self.click_labels = click_labels
        self.value_per_click = value_per_click
        self.N = len(features)
        self.rng = np.random.RandomState(seed)
        
        # Generate synthetic market prices
        self.market_prices = self._generate_market_prices(market_price_config or {})
        
        # Shuffle indices
        self.order = self.rng.permutation(self.N)
        self.position = 0
    
    def _generate_market_prices(self, config):
        """
        Generate synthetic market prices (max competing bids).
        
        Prices are conditioned on features: higher-CTR impressions tend to
        have higher competition (more bidders want them).
        
        Default: log-normal distribution with mean correlated to pCTR.
        """
        base_mean = config.get('base_mean', 20.0)
        ctr_correlation = config.get('ctr_correlation', 10.0)
        noise_std = config.get('noise_std', 0.6)
        price_multiplier = config.get('price_multiplier', 1.0)
        
        N = self.N
        
        # Mean price increases with CTR (popular impressions cost more)
        mean_prices = base_mean + ctr_correlation * self.pctr_true
        mean_prices = np.clip(mean_prices, 1.0, 200.0)
        
        # Log-normal noise
        prices = self.rng.lognormal(
            mean=np.log(mean_prices),
            sigma=noise_std,
            size=N
        )
        
        # Scale
        prices = prices * price_multiplier
        
        # Add feature-dependent variation
        if self.features.shape[1] > 1:
            # Use first two features to add correlated variation
            feature_effect = (
                0.02 * self.features[:, 0] + 
                0.01 * self.features[:, 1]
            )
            prices = prices * np.exp(feature_effect * 0.1)
        
        return np.clip(prices, 0.5, 500.0)
    
    def reset(self):
        """Reset simulator to beginning of auction sequence."""
        self.position = 0
        self.order = self.rng.permutation(self.N)
    
    def next_auction(self):
        """
        Return next auction: (features, pctr, true_click, market_price).
        Returns None when exhausted.
        """
        if self.position >= self.N:
            return None
        
        idx = self.order[self.position]
        self.position += 1
        
        return (
            self.features[idx],
            self.pctr_true[idx],
            int(self.click_labels[idx]),
            self.market_prices[idx]
        )
    
    def simulate_bid(self, bid):
        """
        Simulate the outcome of placing a bid in the current auction.
        
        Must be called after next_auction().
        
        Args:
            bid: Bid price
        
        Returns:
            (won, cost, click_occurred)
        """
        _, _, click, market_price = (
            self.features[self.order[self.position - 1]],
            self.pctr_true[self.order[self.position - 1]],
            int(self.click_labels[self.order[self.position - 1]]),
            self.market_prices[self.order[self.position - 1]]
        )
        
        won = bid >= market_price
        cost = bid if won else 0.0
        click_occurred = click if won else 0
        
        return won, cost, click_occurred, market_price
    
    def run_algorithm(self, algo, ctr_predictor=None):
        """
        Run a bidding algorithm through the full auction sequence.
        
        Args:
            algo: Bidding algorithm instance with .bid(pctr) and .update(won, cost, pctr, d_t)
            ctr_predictor: Optional CTRPredictor. If None, uses self.pctr_true.
        
        Returns:
            results dict with metrics
        """
        self.reset()
        
        # Set budget on algorithm if it supports it
        if hasattr(algo, 'set_budget'):
            algo.set_budget(algo.B if hasattr(algo, 'B') else 5000)
        
        metrics = defaultdict(list)
        total_clicks = 0
        
        while True:
            auction = self.next_auction()
            if auction is None:
                break
            
            features, _, true_click, market_price = auction
            
            # Get CTR prediction
            if ctr_predictor is not None:
                # Use learned CTR model
                pctr = ctr_predictor.predict_single({
                    f'I{i+1}': features[i] for i in range(13)
                } | {
                    f'C{i+1}': features[13+i] for i in range(26)
                })
            else:
                pctr = self.pctr_true[self.order[self.position - 1]]
            
            # Place bid
            bid = algo.bid(pctr, features)
            bid = np.clip(bid, 0, algo.remaining_budget if hasattr(algo, 'remaining_budget') else float('inf'))
            
            # Simulate outcome
            won, cost, click, d_t = self.simulate_bid(bid)
            
            if won:
                total_clicks += click
            
            # Update algorithm
            algo.update(won, cost, pctr, d_t)
            
            # Track metrics
            metrics['bids'].append(bid)
            metrics['won'].append(int(won))
            metrics['cost'].append(cost)
            metrics['pctr'].append(pctr)
            metrics['market_price'].append(market_price)
        
        # Compute summary
        results = algo.get_stats()
        results.update({
            'total_clicks': total_clicks,
            'total_impressions': len(metrics['bids']),
            'total_wins': sum(metrics['won']),
            'total_spent': sum(metrics['cost']),
            'ctr': total_clicks / max(sum(metrics['won']), 1),
            'budget_used_frac': sum(metrics['cost']) / algo.B if hasattr(algo, 'B') else 0,
            'cpc': sum(metrics['cost']) / max(total_clicks, 1),
            'avg_bid': np.mean(metrics['bids']),
            'win_rate': sum(metrics['won']) / max(len(metrics['won']), 1),
            'avg_market_price': np.mean(metrics['market_price']),
        })
        
        return results
    
    def run_comparison(self, algorithms, ctr_predictor=None):
        """
        Run multiple algorithms on the same auction sequence and compare.
        
        Args:
            algorithms: Dict of {name: algorithm_instance}
            ctr_predictor: Shared CTR predictor
        
        Returns:
            Dict of {name: results_dict}
        """
        all_results = {}
        
        for name, algo in algorithms.items():
            print(f"\nRunning {name}...")
            results = self.run_algorithm(algo, ctr_predictor)
            all_results[name] = results
            
            print(f"  Clicks: {results['total_clicks']}")
            print(f"  Spend: {results['total_spent']:.2f}")
            print(f"  Budget used: {results.get('budget_used_frac', 0):.1%}")
            print(f"  CPC: {results.get('cpc', 0):.2f}")
        
        return all_results