File size: 18,310 Bytes
50c7ef1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
68f2a01
50c7ef1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
68f2a01
50c7ef1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
68f2a01
50c7ef1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
"""

Transparent AI Reasoning Engine for POLYMEROS

Provides explainable predictions with uncertainty quantification and hypothesis generation

"""

import numpy as np
import torch
import torch.nn.functional as F
from typing import Dict, List, Any, Tuple, Optional
from dataclasses import dataclass
import warnings

try:
    import shap

    SHAP_AVAILABLE = True
except ImportError:
    SHAP_AVAILABLE = False
    warnings.warn("SHAP not available. Install with: pip install shap")


@dataclass
class PredictionExplanation:
    """Comprehensive explanation for a model prediction"""

    prediction: int
    confidence: float
    confidence_level: str
    probabilities: np.ndarray
    feature_importance: Dict[str, float]
    reasoning_chain: List[str]
    uncertainty_sources: List[str]
    similar_cases: List[Dict[str, Any]]
    confidence_intervals: Dict[str, Tuple[float, float]]


@dataclass
class Hypothesis:
    """AI-generated scientific hypothesis"""

    statement: str
    confidence: float
    supporting_evidence: List[str]
    testable_predictions: List[str]
    suggested_experiments: List[str]
    related_literature: List[str]


class UncertaintyEstimator:
    """Bayesian uncertainty estimation for model predictions"""

    def __init__(self, model, n_samples: int = 100):
        self.model = model
        self.n_samples = n_samples
        self.epistemic_uncertainty = None
        self.aleatoric_uncertainty = None

    def estimate_uncertainty(self, x: torch.Tensor) -> Dict[str, float]:
        """Estimate prediction uncertainty using Monte Carlo dropout"""
        self.model.train()  # Enable dropout

        predictions = []
        with torch.no_grad():
            for _ in range(self.n_samples):
                pred = F.softmax(self.model(x), dim=1)
                predictions.append(pred.cpu().numpy())

        predictions = np.array(predictions)

        # Calculate uncertainties
        mean_pred = np.mean(predictions, axis=0)
        epistemic = np.var(predictions, axis=0)  # Model uncertainty
        aleatoric = np.mean(predictions * (1 - predictions), axis=0)  # Data uncertainty

        total_uncertainty = epistemic + aleatoric

        return {
            "epistemic": float(np.mean(epistemic)),
            "aleatoric": float(np.mean(aleatoric)),
            "total": float(np.mean(total_uncertainty)),
            "prediction_variance": float(np.var(mean_pred)),
        }

    def confidence_intervals(

        self, x: torch.Tensor, confidence_level: float = 0.95

    ) -> Dict[str, Tuple[float, float]]:
        """Calculate confidence intervals for predictions"""
        self.model.train()

        predictions = []
        with torch.no_grad():
            for _ in range(self.n_samples):
                pred = F.softmax(self.model(x), dim=1)
                predictions.append(pred.cpu().numpy().flatten())

        predictions = np.array(predictions)

        alpha = 1 - confidence_level
        lower_percentile = (alpha / 2) * 100
        upper_percentile = (1 - alpha / 2) * 100

        intervals = {}
        for i in range(predictions.shape[1]):
            lower = np.percentile(predictions[:, i], lower_percentile)
            upper = np.percentile(predictions[:, i], upper_percentile)
            intervals[f"class_{i}"] = (lower, upper)

        return intervals


class FeatureImportanceAnalyzer:
    """Advanced feature importance analysis for spectral data"""

    def __init__(self, model):
        self.model = model
        self.shap_explainer = None

        if SHAP_AVAILABLE:
            try:
                # Initialize SHAP explainer for the model
                if SHAP_AVAILABLE:
                    if SHAP_AVAILABLE:
                        self.shap_explainer = shap.DeepExplainer(  # type: ignore
                            model, torch.zeros(1, 500)
                        )
                    else:
                        self.shap_explainer = None
                else:
                    self.shap_explainer = None
            except (ValueError, RuntimeError) as e:
                warnings.warn(f"Could not initialize SHAP explainer: {e}")

    def analyze_feature_importance(

        self, x: torch.Tensor, wavenumbers: Optional[np.ndarray] = None

    ) -> Dict[str, Any]:
        """Comprehensive feature importance analysis"""
        importance_data = {}

        # SHAP analysis (if available)
        if self.shap_explainer is not None:
            try:
                shap_values = self.shap_explainer.shap_values(x)
                importance_data["shap_values"] = shap_values
                importance_data["shap_available"] = True
            except (ValueError, RuntimeError) as e:
                warnings.warn(f"SHAP analysis failed: {e}")
                importance_data["shap_available"] = False
        else:
            importance_data["shap_available"] = False

        # Gradient-based importance
        x.requires_grad_(True)
        self.model.eval()

        output = self.model(x)
        predicted_class = torch.argmax(output, dim=1)

        # Calculate gradients
        self.model.zero_grad()
        output[0, predicted_class].backward()

        if x.grad is not None:
            gradients = x.grad.detach().abs().cpu().numpy().flatten()
        else:
            raise RuntimeError(
                "Gradients were not computed. Ensure x.requires_grad_(True) is set correctly."
            )

        importance_data["gradient_importance"] = gradients

        # Integrated gradients approximation
        integrated_grads = self._integrated_gradients(x, predicted_class)
        importance_data["integrated_gradients"] = integrated_grads

        # Spectral region importance
        if wavenumbers is not None:
            region_importance = self._analyze_spectral_regions(gradients, wavenumbers)
            importance_data["spectral_regions"] = region_importance

        return importance_data

    def _integrated_gradients(

        self, x: torch.Tensor, target_class: torch.Tensor, steps: int = 50

    ) -> np.ndarray:
        """Calculate integrated gradients for feature importance"""
        baseline = torch.zeros_like(x)

        integrated_grads = np.zeros(x.shape[1])

        for i in range(steps):
            alpha = i / steps
            interpolated = baseline + alpha * (x - baseline)
            interpolated.requires_grad_(True)

            output = self.model(interpolated)
            self.model.zero_grad()
            output[0, target_class].backward(retain_graph=True)

            if interpolated.grad is not None:
                grads = interpolated.grad.cpu().numpy().flatten()
                integrated_grads += grads

        integrated_grads = (
            integrated_grads * (x - baseline).detach().cpu().numpy().flatten() / steps
        )
        return integrated_grads

    def _analyze_spectral_regions(

        self, importance: np.ndarray, wavenumbers: np.ndarray

    ) -> Dict[str, float]:
        """Analyze importance by common spectral regions"""
        regions = {
            "fingerprint": (400, 1500),
            "ch_stretch": (2800, 3100),
            "oh_stretch": (3200, 3700),
            "carbonyl": (1600, 1800),
            "aromatic": (1450, 1650),
        }

        region_importance = {}

        for region_name, (low, high) in regions.items():
            mask = (wavenumbers >= low) & (wavenumbers <= high)
            if np.any(mask):
                region_importance[region_name] = float(np.mean(importance[mask]))
            else:
                region_importance[region_name] = 0.0

        return region_importance


class HypothesisGenerator:
    """AI-driven scientific hypothesis generation"""

    def __init__(self):
        self.hypothesis_templates = [
            "The spectral differences in the {region} region suggest {mechanism} as a primary degradation pathway",
            "Enhanced intensity at {wavenumber} cm⁻¹ indicates {chemical_change} in weathered samples",
            "The correlation between {feature1} and {feature2} suggests {relationship}",
            "Baseline shifts in {region} region may indicate {structural_change}",
        ]

    def generate_hypotheses(

        self, explanation: PredictionExplanation

    ) -> List[Hypothesis]:
        """Generate testable hypotheses based on model predictions and explanations"""
        hypotheses = []

        # Analyze feature importance for hypothesis generation
        important_features = self._identify_key_features(explanation.feature_importance)

        for feature_info in important_features:
            hypothesis = self._generate_single_hypothesis(feature_info, explanation)
            if hypothesis:
                hypotheses.append(hypothesis)

        return hypotheses

    def _identify_key_features(

        self, feature_importance: Dict[str, float]

    ) -> List[Dict[str, Any]]:
        """Identify key features for hypothesis generation"""
        # Sort features by importance
        sorted_features = sorted(
            feature_importance.items(), key=lambda x: abs(x[1]), reverse=True
        )

        key_features = []
        for feature_name, importance in sorted_features[:5]:  # Top 5 features
            feature_info = {
                "name": feature_name,
                "importance": importance,
                "type": self._classify_feature_type(feature_name),
                "chemical_significance": self._get_chemical_significance(feature_name),
            }
            key_features.append(feature_info)

        return key_features

    def _classify_feature_type(self, feature_name: str) -> str:
        """Classify spectral feature type"""
        if "fingerprint" in feature_name.lower():
            return "fingerprint"
        elif "stretch" in feature_name.lower():
            return "vibrational"
        elif "carbonyl" in feature_name.lower():
            return "functional_group"
        else:
            return "general"

    def _get_chemical_significance(self, feature_name: str) -> str:
        """Get chemical significance of spectral feature"""
        significance_map = {
            "fingerprint": "molecular backbone structure",
            "ch_stretch": "aliphatic chain integrity",
            "oh_stretch": "hydrogen bonding and hydration",
            "carbonyl": "oxidative degradation products",
            "aromatic": "aromatic ring preservation",
        }

        for key, significance in significance_map.items():
            if key in feature_name.lower():
                return significance

        return "structural changes"

    def _generate_single_hypothesis(

        self, feature_info: Dict[str, Any], explanation: PredictionExplanation

    ) -> Optional[Hypothesis]:
        """Generate a single hypothesis from feature information"""
        if feature_info["importance"] < 0.1:  # Skip low-importance features
            return None

        # Create hypothesis statement
        statement = f"Changes in {feature_info['name']} region indicate {feature_info['chemical_significance']} during polymer weathering"

        # Generate supporting evidence
        evidence = [
            f"Feature importance score: {feature_info['importance']:.3f}",
            f"Classification confidence: {explanation.confidence:.3f}",
            f"Chemical significance: {feature_info['chemical_significance']}",
        ]

        # Generate testable predictions
        predictions = [
            f"Controlled weathering experiments should show progressive changes in {feature_info['name']} region",
            f"Different polymer types should exhibit varying {feature_info['name']} responses to weathering",
        ]

        # Suggest experiments
        experiments = [
            f"Time-series weathering study monitoring {feature_info['name']} region",
            f"Comparative analysis across polymer types focusing on {feature_info['chemical_significance']}",
            "Cross-validation with other analytical techniques (DSC, GPC, etc.)",
        ]

        return Hypothesis(
            statement=statement,
            confidence=min(0.9, feature_info["importance"] * explanation.confidence),
            supporting_evidence=evidence,
            testable_predictions=predictions,
            suggested_experiments=experiments,
            related_literature=[],  # Could be populated with literature search
        )


class TransparentAIEngine:
    """Main transparent AI engine combining all reasoning components"""

    def __init__(self, model):
        self.model = model
        self.uncertainty_estimator = UncertaintyEstimator(model)
        self.feature_analyzer = FeatureImportanceAnalyzer(model)
        self.hypothesis_generator = HypothesisGenerator()

    def predict_with_explanation(

        self, x: torch.Tensor, wavenumbers: Optional[np.ndarray] = None

    ) -> PredictionExplanation:
        """Generate comprehensive prediction with full explanation"""
        self.model.eval()

        # Get basic prediction
        with torch.no_grad():
            logits = self.model(x)
            probabilities = F.softmax(logits, dim=1).cpu().numpy().flatten()
            prediction = int(torch.argmax(logits, dim=1).item())
            confidence = float(np.max(probabilities))

        # Determine confidence level
        if confidence >= 0.80:
            confidence_level = "HIGH"
        elif confidence >= 0.60:
            confidence_level = "MEDIUM"
        else:
            confidence_level = "LOW"

        # Get uncertainty estimation
        uncertainties = self.uncertainty_estimator.estimate_uncertainty(x)
        confidence_intervals = self.uncertainty_estimator.confidence_intervals(x)

        # Analyze feature importance
        importance_data = self.feature_analyzer.analyze_feature_importance(
            x, wavenumbers
        )

        # Create feature importance dictionary
        if wavenumbers is not None and "spectral_regions" in importance_data:
            feature_importance = importance_data["spectral_regions"]
        else:
            # Use gradient importance
            gradients = importance_data.get("gradient_importance", [])
            feature_importance = {
                f"feature_{i}": float(val) for i, val in enumerate(gradients[:10])
            }

        # Generate reasoning chain
        reasoning_chain = self._generate_reasoning_chain(
            prediction, confidence, feature_importance, uncertainties
        )

        # Identify uncertainty sources
        uncertainty_sources = self._identify_uncertainty_sources(uncertainties)

        # Create explanation object
        explanation = PredictionExplanation(
            prediction=prediction,
            confidence=confidence,
            confidence_level=confidence_level,
            probabilities=probabilities,
            feature_importance=feature_importance,
            reasoning_chain=reasoning_chain,
            uncertainty_sources=uncertainty_sources,
            similar_cases=[],  # Could be populated with case-based reasoning
            confidence_intervals=confidence_intervals,
        )

        return explanation

    def generate_hypotheses(

        self, explanation: PredictionExplanation

    ) -> List[Hypothesis]:
        """Generate scientific hypotheses based on prediction explanation"""
        return self.hypothesis_generator.generate_hypotheses(explanation)

    def _generate_reasoning_chain(

        self,

        prediction: int,

        confidence: float,

        feature_importance: Dict[str, float],

        uncertainties: Dict[str, float],

    ) -> List[str]:
        """Generate human-readable reasoning chain"""
        reasoning = []

        # Start with prediction
        class_names = ["Stable", "Weathered"]
        reasoning.append(
            f"Model predicts: {class_names[prediction]} (confidence: {confidence:.3f})"
        )

        # Add feature analysis
        top_features = sorted(
            feature_importance.items(), key=lambda x: abs(x[1]), reverse=True
        )[:3]

        for feature, importance in top_features:
            reasoning.append(
                f"Key evidence: {feature} region shows importance score {importance:.3f}"
            )

        # Add uncertainty analysis
        total_uncertainty = uncertainties.get("total", 0)
        if total_uncertainty > 0.1:
            reasoning.append(
                f"High uncertainty detected ({total_uncertainty:.3f}) - suggests ambiguous case"
            )

        # Add confidence assessment
        if confidence > 0.8:
            reasoning.append(
                "High confidence: Strong spectral signature for classification"
            )
        elif confidence > 0.6:
            reasoning.append("Medium confidence: Some ambiguity in spectral features")
        else:
            reasoning.append("Low confidence: Weak or conflicting spectral evidence")

        return reasoning

    def _identify_uncertainty_sources(

        self, uncertainties: Dict[str, float]

    ) -> List[str]:
        """Identify sources of prediction uncertainty"""
        sources = []

        epistemic = uncertainties.get("epistemic", 0)
        aleatoric = uncertainties.get("aleatoric", 0)

        if epistemic > 0.05:
            sources.append(
                "Model uncertainty: Limited training data for this type of spectrum"
            )

        if aleatoric > 0.05:
            sources.append("Data uncertainty: Noisy or degraded spectral quality")

        if uncertainties.get("prediction_variance", 0) > 0.1:
            sources.append("Prediction instability: Multiple possible interpretations")

        if not sources:
            sources.append("Low uncertainty: Clear and unambiguous classification")

        return sources