local-deep-research 0.4.4__py3-none-any.whl → 0.5.0__py3-none-any.whl

local_deep_research/advanced_search_system/strategies/smart_decomposition_strategy.py

@@ -0,0 +1,284 @@
+"""
+Smart Decomposition Strategy that chooses between recursive and adaptive approaches.
+
+This strategy analyzes the query type and chooses the most appropriate
+decomposition method for optimal results.
+"""
+
+from enum import Enum
+from typing import Any, Dict, List
+
+from langchain_core.language_models import BaseChatModel
+from loguru import logger
+
+from ...utilities.search_utilities import remove_think_tags
+from .adaptive_decomposition_strategy import AdaptiveDecompositionStrategy
+from .base_strategy import BaseSearchStrategy
+from .browsecomp_optimized_strategy import BrowseCompOptimizedStrategy
+from .evidence_based_strategy import EvidenceBasedStrategy
+from .recursive_decomposition_strategy import RecursiveDecompositionStrategy
+
+
+class QueryType(Enum):
+    """Types of queries that benefit from different strategies."""
+
+    PUZZLE_LIKE = "puzzle_like"  # Queries with specific clues/constraints
+    HIERARCHICAL = "hierarchical"  # Multi-part questions with clear structure
+    EXPLORATORY = "exploratory"  # Open-ended research questions
+    COMPARATIVE = "comparative"  # Comparison between multiple entities
+    FACTUAL = "factual"  # Simple factual questions
+    CONSTRAINT_BASED = (
+        "constraint_based"  # Queries with multiple verifiable constraints
+    )
+
+
+class SmartDecompositionStrategy(BaseSearchStrategy):
+    """
+    A meta-strategy that intelligently chooses between decomposition approaches.
+
+    It analyzes the query characteristics and selects either:
+    - Adaptive decomposition for puzzle-like or constraint-based queries
+    - Recursive decomposition for clearly hierarchical queries
+    - Direct search for simple factual queries
+    """
+
+    def __init__(
+        self,
+        model: BaseChatModel,
+        search: Any,
+        all_links_of_system: List[str],
+        **kwargs,
+    ):
+        """Initialize the smart decomposition strategy.
+
+        Args:
+            model: The language model to use
+            search: The search engine instance
+            all_links_of_system: List to store all encountered links
+            **kwargs: Additional parameters for sub-strategies
+        """
+        super().__init__(all_links_of_system)
+        self.model = model
+        self.search = search
+        self.strategy_params = kwargs
+
+    def analyze_topic(self, query: str) -> Dict:
+        """Analyze a topic using the most appropriate strategy.
+
+        Args:
+            query: The research query to analyze
+
+        Returns:
+            Dictionary containing analysis results
+        """
+        # Determine query type
+        query_type = self._classify_query(query)
+        logger.info(f"Query classified as: {query_type.value}")
+
+        # Progress callback for UI
+        if self.progress_callback:
+            self.progress_callback(
+                f"Analyzing query type: {query_type.value}",
+                5,
+                {
+                    "phase": "query_classification",
+                    "query_type": query_type.value,
+                },
+            )
+
+        # Choose and execute appropriate strategy
+        if query_type == QueryType.PUZZLE_LIKE:
+            # Check if it's specifically a BrowseComp-style puzzle
+            if self._is_browsecomp_style(query):
+                return self._use_browsecomp_strategy(query)
+            else:
+                return self._use_evidence_strategy(query)
+        elif query_type == QueryType.CONSTRAINT_BASED:
+            return self._use_evidence_strategy(query)
+        elif query_type == QueryType.HIERARCHICAL:
+            return self._use_recursive_strategy(query)
+        elif query_type in [QueryType.COMPARATIVE, QueryType.EXPLORATORY]:
+            return self._use_recursive_strategy(query)
+        else:  # FACTUAL or unknown
+            return self._use_adaptive_strategy(query)
+
+    def _classify_query(self, query: str) -> QueryType:
+        """Classify the query type to determine best strategy.
+
+        Args:
+            query: The query to classify
+
+        Returns:
+            QueryType enum value
+        """
+        prompt = f"""Analyze this query and classify its type:
+
+Query: {query}
+
+Query Types:
+1. PUZZLE_LIKE - Contains specific clues, constraints, or puzzle elements that need to be solved step by step
+2. HIERARCHICAL - Multi-part question with clear subtopics that can be addressed independently
+3. EXPLORATORY - Open-ended research question seeking broad understanding
+4. COMPARATIVE - Requires comparing multiple entities or concepts
+5. FACTUAL - Simple factual question with a direct answer
+6. CONSTRAINT_BASED - Has multiple specific constraints that need verification (dates, numbers, properties)
+
+Consider:
+- Does it have specific constraints or clues to follow?
+- Can it be broken into independent subtasks?
+- Is it seeking a specific answer or broad understanding?
+- Does it require step-by-step verification?
+
+Respond with:
+QUERY_TYPE: [type]
+REASONING: [brief explanation]
+KEY_CHARACTERISTICS: [list main characteristics]
+"""
+
+        response = self.model.invoke(prompt)
+        content = remove_think_tags(response.content)
+
+        # Parse response
+        query_type = QueryType.EXPLORATORY  # Default
+
+        lines = content.strip().split("\n")
+        for line in lines:
+            if line.startswith("QUERY_TYPE:"):
+                type_str = line.split(":", 1)[1].strip()
+                try:
+                    query_type = QueryType[type_str]
+                except KeyError:
+                    # Try to match by value
+                    for q_type in QueryType:
+                        if q_type.value in type_str.lower():
+                            query_type = q_type
+                            break
+
+        logger.info(f"Query classification reasoning: {content}")
+        return query_type
+
+    def _use_adaptive_strategy(self, query: str) -> Dict:
+        """Use adaptive decomposition strategy.
+
+        Args:
+            query: The query to analyze
+
+        Returns:
+            Analysis results
+        """
+        logger.info("Using adaptive decomposition strategy")
+
+        strategy = AdaptiveDecompositionStrategy(
+            model=self.model,
+            search=self.search,
+            all_links_of_system=self.all_links_of_system,
+            **self.strategy_params,
+        )
+
+        if self.progress_callback:
+            strategy.set_progress_callback(self.progress_callback)
+
+        return strategy.analyze_topic(query)
+
+    def _is_browsecomp_style(self, query: str) -> bool:
+        """Check if query is a BrowseComp-style puzzle.
+
+        Args:
+            query: The query to check
+
+        Returns:
+            True if it appears to be a BrowseComp puzzle
+        """
+        browsecomp_indicators = [
+            "specific scenic location",
+            "body part",
+            "fell from",
+            "viewpoint",
+            "search and rescue",
+            "SAR incidents",
+            "exact answer",
+            "multiple clues",
+            "last ice age",
+            "times more",
+            "between",
+            "what is the name",
+        ]
+
+        lower_query = query.lower()
+        indicator_count = sum(
+            1
+            for indicator in browsecomp_indicators
+            if indicator.lower() in lower_query
+        )
+
+        # If 3+ indicators present, likely BrowseComp-style
+        return indicator_count >= 3
+
+    def _use_browsecomp_strategy(self, query: str) -> Dict:
+        """Use BrowseComp-optimized strategy.
+
+        Args:
+            query: The query to analyze
+
+        Returns:
+            Analysis results
+        """
+        logger.info("Using BrowseComp-optimized strategy for puzzle query")
+
+        strategy = BrowseCompOptimizedStrategy(
+            model=self.model,
+            search=self.search,
+            all_links_of_system=self.all_links_of_system,
+            **self.strategy_params,
+        )
+
+        if self.progress_callback:
+            strategy.set_progress_callback(self.progress_callback)
+
+        return strategy.analyze_topic(query)
+
+    def _use_evidence_strategy(self, query: str) -> Dict:
+        """Use evidence-based strategy.
+
+        Args:
+            query: The query to analyze
+
+        Returns:
+            Analysis results
+        """
+        logger.info("Using evidence-based strategy for constraint-based query")
+
+        strategy = EvidenceBasedStrategy(
+            model=self.model,
+            search=self.search,
+            all_links_of_system=self.all_links_of_system,
+            **self.strategy_params,
+        )
+
+        if self.progress_callback:
+            strategy.set_progress_callback(self.progress_callback)
+
+        return strategy.analyze_topic(query)
+
+    def _use_recursive_strategy(self, query: str) -> Dict:
+        """Use recursive decomposition strategy.
+
+        Args:
+            query: The query to analyze
+
+        Returns:
+            Analysis results
+        """
+        logger.info("Using recursive decomposition strategy")
+
+        strategy = RecursiveDecompositionStrategy(
+            model=self.model,
+            search=self.search,
+            all_links_of_system=self.all_links_of_system,
+            **self.strategy_params,
+        )
+
+        if self.progress_callback:
+            strategy.set_progress_callback(self.progress_callback)
+
+        return strategy.analyze_topic(query)