web_interface/server.py

#!/usr/bin/env python3
"""
Flask server for POA Graph Web Interface
"""

import glob
import os
import pickle
import re
import sys

from flask import Flask, jsonify, request, send_from_directory
from flask_cors import CORS

# Get the repository root directory (parent of web_interface)
REPO_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))

# Add the repository root to the path so we can import the POA graph modules
sys.path.append(REPO_ROOT)

from src.new_text_alignment import TextSeqGraphAlignment
from src.text_poa_graph import TextPOAGraph

try:
    from src.generation_methods import decode_consensus
except ImportError:
    decode_consensus = None

app = Flask(__name__)
CORS(app)  # Enable CORS for all routes

# Base paths for different datasets (relative to repo root)
GRAPH_PATHS = {
    "bio": os.path.join(REPO_ROOT, "results/graphs/HALoGEN/bio"),
    "fp": os.path.join(REPO_ROOT, "results/graphs/HALoGEN/fp"),
    "hist": os.path.join(REPO_ROOT, "results/graphs/HALoGEN/hist"),
    "refs": os.path.join(REPO_ROOT, "results/graphs/HALoGEN/refs"),
    "math": os.path.join(REPO_ROOT, "results/graphs/HALoGEN/MATH"),
    "aime": os.path.join(REPO_ROOT, "results/graphs/HALoGEN/AIME"),
}

MODELS = ["qwen72b", "qwen7b", "llama8b", "llama70b", "olmo7b", "olmo32b"]


@app.route("/")
def index():
    """Serve the main HTML file"""
    return send_from_directory(".", "index.html")


@app.route("/api/datasets", methods=["GET"])
def get_datasets():
    """Get available datasets"""
    datasets = []
    for dataset_name, path in GRAPH_PATHS.items():
        if os.path.exists(path):
            # Count available graphs
            pkl_files = glob.glob(os.path.join(path, "*.pkl"))
            datasets.append(
                {
                    "name": dataset_name,
                    "display_name": dataset_name.upper(),
                    "path": path,
                    "count": len(pkl_files),
                }
            )
    return jsonify({"datasets": datasets})


@app.route("/api/models", methods=["GET"])
def get_models():
    """Get available models for a specific entity"""
    entity = request.args.get("entity")
    dataset = request.args.get("dataset")

    if not entity:
        return jsonify({"error": "Entity parameter required"}), 400

    if not dataset or dataset not in GRAPH_PATHS:
        return jsonify({"error": "Invalid dataset"}), 400

    path = GRAPH_PATHS[dataset]
    if not os.path.exists(path):
        return jsonify({"error": "Dataset path not found"}), 404

    models = []
    pkl_files = glob.glob(os.path.join(path, "*.pkl"))

    for pkl_file in pkl_files:
        filename = os.path.basename(pkl_file)

        # Different filename patterns for different datasets
        if dataset == "bio":
            # Format: bio_graph_{entity}_merged_{model}.pkl
            match = re.match(r"bio_graph_(.+?)_merged_(\w+)\.pkl", filename)
            if match:
                entity_name, model = match.groups()
                if entity_name == entity:
                    models.append({"model": model, "filename": filename, "filepath": pkl_file})
        elif dataset == "fp":
            # Format: fp_graph_{number}_merged_{model}.pkl
            match = re.match(r"fp_graph_(\d+)_merged_(\w+)\.pkl", filename)
            if match:
                entity_name, model = match.groups()
                if f"Problem {entity_name}" == entity:
                    models.append({"model": model, "filename": filename, "filepath": pkl_file})
        elif dataset == "math":
            # Format: qwen72_math_{number}.pkl
            match = re.match(r"qwen72_math_(\d+)\.pkl", filename)
            if match:
                entity_name = match.group(1)
                if f"Math Problem {entity_name}" == entity:
                    models.append({"model": "qwen72b", "filename": filename, "filepath": pkl_file})
        elif dataset == "aime":
            # Format: aime_qwen72b_{number}.pkl
            match = re.match(r"aime_qwen72b_(\d+)\.pkl", filename)
            if match:
                entity_name = match.group(1)
                if f"AIME Problem {entity_name}" == entity:
                    models.append({"model": "qwen72b", "filename": filename, "filepath": pkl_file})
        else:
            # Generic pattern for other datasets
            match = re.match(r"(\w+)_graph_(.+?)_merged_(\w+)\.pkl", filename)
            if match:
                task, entity_name, model = match.groups()
                if entity_name == entity:
                    models.append({"model": model, "filename": filename, "filepath": pkl_file})

    return jsonify({"models": models})


@app.route("/api/entities", methods=["GET"])
def get_entities():
    """Get available entities for a dataset"""
    dataset = request.args.get("dataset")
    if not dataset or dataset not in GRAPH_PATHS:
        return jsonify({"error": "Invalid dataset"}), 400

    path = GRAPH_PATHS[dataset]
    if not os.path.exists(path):
        return jsonify({"error": "Dataset path not found"}), 404

    entities = []
    pkl_files = glob.glob(os.path.join(path, "*.pkl"))

    for pkl_file in pkl_files:
        filename = os.path.basename(pkl_file)

        # Different filename patterns for different datasets
        if dataset == "bio":
            # Format: bio_graph_{entity}_merged_{model}.pkl
            match = re.match(r"bio_graph_(.+?)_merged_(\w+)\.pkl", filename)
            if match:
                entity_name, model = match.groups()
                entities.append(
                    {
                        "entity": entity_name,
                        "model": model,
                        "filename": filename,
                        "filepath": pkl_file,
                    }
                )
        elif dataset == "fp":
            # Format: fp_graph_{number}_merged_{model}.pkl
            match = re.match(r"fp_graph_(\d+)_merged_(\w+)\.pkl", filename)
            if match:
                entity_name, model = match.groups()
                entities.append(
                    {
                        "entity": f"Problem {entity_name}",
                        "model": model,
                        "filename": filename,
                        "filepath": pkl_file,
                    }
                )
        elif dataset == "math":
            # Format: qwen72_math_{number}.pkl
            match = re.match(r"qwen72_math_(\d+)\.pkl", filename)
            if match:
                entity_name = match.group(1)
                entities.append(
                    {
                        "entity": f"Math Problem {entity_name}",
                        "model": "qwen72b",
                        "filename": filename,
                        "filepath": pkl_file,
                    }
                )
        elif dataset == "aime":
            # Format: aime_qwen72b_{number}.pkl
            match = re.match(r"aime_qwen72b_(\d+)\.pkl", filename)
            if match:
                entity_name = match.group(1)
                entities.append(
                    {
                        "entity": f"AIME Problem {entity_name}",
                        "model": "qwen72b",
                        "filename": filename,
                        "filepath": pkl_file,
                    }
                )
        else:
            # Generic pattern for other datasets
            match = re.match(r"(\w+)_graph_(.+?)_merged_(\w+)\.pkl", filename)
            if match:
                task, entity_name, model = match.groups()
                entities.append(
                    {
                        "entity": entity_name,
                        "model": model,
                        "filename": filename,
                        "filepath": pkl_file,
                    }
                )

    return jsonify({"entities": entities})


@app.route("/api/load_existing_graph", methods=["POST"])
def load_existing_graph():
    """Load an existing graph from the stored pickle files"""
    try:
        data = request.get_json()
        filepath = data.get("filepath")

        if not filepath or not os.path.exists(filepath):
            return jsonify({"error": "Graph file not found"}), 404

        # Read and load the pickle file
        try:
            with open(filepath, "rb") as f:
                graph = pickle.load(f)
        except Exception as e:
            return jsonify({"error": f"Error loading pickle file: {str(e)}"}), 500

        if not isinstance(graph, TextPOAGraph):
            return jsonify({"error": "File does not contain a valid POA graph"}), 400

        # Convert to JSON format for vis.js
        nodes = []
        edges = []

        try:
            # Get consensus nodes for coloring
            consensus_nodes = set(graph.consensus_node_ids)

            # Create nodes using the same logic as jsOutput
            for node in graph.nodeiterator()():
                title_text = ""
                if node.sequences:
                    title_text += f"Sequences: {node.sequences}"
                if node.variations:
                    title_text += ";;;".join(
                        [f"{sequence_id}: {text}" for sequence_id, text in node.variations.items()]
                    )
                    title_text = title_text.replace('"', "'")

                # Use the same color logic as jsOutput
                color = "#ceeab2" if node.ID in consensus_nodes else "#cae0e6"

                node_data = {
                    "id": node.ID,
                    "label": f"{node.ID}: {node.text}",
                    "title": title_text,
                    "color": color,
                }
                nodes.append(node_data)

            # Create edges using the same logic as jsOutput
            for node in graph.nodeiterator()():
                nodeID = node.ID  # Keep as integer
                for edge in node.outEdges:
                    target = edge  # Keep as integer
                    weight = node.outEdges[edge].weight + 1.5
                    edge_data = {
                        "from": nodeID,
                        "to": target,
                        "value": weight,
                        "color": "#cae0e6",
                        "arrows": "to",
                    }
                    edges.append(edge_data)
        except Exception as e:
            return jsonify({"error": f"Error processing graph data: {str(e)}"}), 500

        # Extract metadata from filename
        filename = os.path.basename(filepath)
        metadata = {}

        try:
            # Different filename patterns for different datasets
            if filename.startswith("bio_graph_"):
                # Format: bio_graph_{entity}_merged_{model}.pkl
                match = re.match(r"bio_graph_(.+?)_merged_(\w+)\.pkl", filename)
                if match:
                    entity_name, model = match.groups()
                    metadata = {
                        "task": "bio",
                        "entity": entity_name,
                        "model": model,
                        "filename": filename,
                    }
            elif filename.startswith("fp_graph_"):
                # Format: fp_graph_{number}_merged_{model}.pkl
                match = re.match(r"fp_graph_(\d+)_merged_(\w+)\.pkl", filename)
                if match:
                    entity_name, model = match.groups()
                    metadata = {
                        "task": "fp",
                        "entity": f"Problem {entity_name}",
                        "model": model,
                        "filename": filename,
                    }
            elif filename.startswith("qwen72_math_"):
                # Format: qwen72_math_{number}.pkl
                match = re.match(r"qwen72_math_(\d+)\.pkl", filename)
                if match:
                    entity_name = match.group(1)
                    metadata = {
                        "task": "math",
                        "entity": f"Math Problem {entity_name}",
                        "model": "qwen72b",
                        "filename": filename,
                    }
            elif filename.startswith("aime_qwen72b_"):
                # Format: aime_qwen72b_{number}.pkl
                match = re.match(r"aime_qwen72b_(\d+)\.pkl", filename)
                if match:
                    entity_name = match.group(1)
                    metadata = {
                        "task": "aime",
                        "entity": f"AIME Problem {entity_name}",
                        "model": "qwen72b",
                        "filename": filename,
                    }
            else:
                # Generic pattern for other datasets
                match = re.match(r"(\w+)_graph_(.+?)_merged_(\w+)\.pkl", filename)
                if match:
                    task, entity_name, model = match.groups()
                    metadata = {
                        "task": task,
                        "entity": entity_name,
                        "model": model,
                        "filename": filename,
                    }
        except Exception:
            # Don't fail the request if metadata extraction fails
            pass

        # Extract text from consensus nodes
        consensus_text = ""
        try:
            consensus_nodes = set(graph.consensus_node_ids)
            consensus_node_texts = []
            for node in graph.nodeiterator()():
                if node.ID in consensus_nodes and node.text and node.text.strip():
                    consensus_node_texts.append(node.text.strip())
            consensus_text = " ".join(consensus_node_texts)
        except Exception:
            consensus_text = ""

        # Check if we should compute consensus using decode_consensus
        compute_consensus = data.get("compute_consensus", False)
        if compute_consensus and decode_consensus:
            try:
                # Determine task from metadata or default to "bio"
                task = metadata.get("task", "bio") if metadata else "bio"
                consensus_text = decode_consensus(graph, selection_threshold=0.5, task=task)
            except Exception as e:
                print(f"DEBUG: Error computing consensus with decode_consensus: {e}")
                # Keep the original consensus text if decode_consensus fails

        # Get original sequences
        try:
            raw_sequences = graph._seqs if hasattr(graph, "_seqs") else []
            # Process sequences: join with spaces and remove "||"
            print(f"DEBUG: Raw sequences: {raw_sequences}")
            original_sequences = []
            for seq in raw_sequences:
                if isinstance(seq, list):
                    # Join list elements with spaces
                    processed_seq = " ".join(str(item) for item in seq)
                else:
                    processed_seq = str(seq)
                # Remove "||" characters
                processed_seq = processed_seq.replace("||", "")
                print(f"DEBUG: Processed sequence: {processed_seq}")
                original_sequences.append(processed_seq)
        except Exception:
            original_sequences = []

        result = {
            "success": True,
            "nodes": nodes,
            "edges": edges,
            "num_sequences": graph.num_sequences,
            "num_nodes": len(nodes),
            "num_edges": len(edges),
            "metadata": metadata,
            "consensus_text": consensus_text,
            "original_sequences": original_sequences,
        }

        return jsonify(result)

    except Exception as e:
        return jsonify({"error": str(e)}), 500


@app.route("/api/create_graph", methods=["POST"])
def create_graph():
    """Create a POA graph from text sequences"""
    try:
        data = request.get_json()
        sequences = data.get("sequences", [])

        if len(sequences) < 2:
            return jsonify({"error": "At least 2 sequences are required"}), 400

        print(f"DEBUG: Creating graph with sequences: {sequences}")

        # Create the graph with first sequence as string
        graph = TextPOAGraph(sequences[0], label=0)
        print("DEBUG: Initial graph created")

        # Add remaining sequences
        for i, sequence in enumerate(sequences[1:], 1):
            print(f"DEBUG: Adding sequence {i}: {sequence}")
            alignment = TextSeqGraphAlignment(
                text=sequence,
                graph=graph,
                fastMethod=True,
                globalAlign=True,
                matchscore=1,
                mismatchscore=-2,
                gap_open=-1,
            )
            graph.incorporateSeqAlignment(alignment, sequence, label=i)

        print("DEBUG: All sequences added")

        # Refine the graph with proper domain and model parameters
        graph.refine_graph(verbose=False, domain="text", model="gpt-4o-mini")
        print("DEBUG: Graph refined")

        # Convert to JSON format for vis.js
        nodes = []
        edges = []

        try:
            print("DEBUG: Starting to process graph data")
            # Get consensus nodes for coloring (make it optional)
            try:
                consensus_nodes = set(graph.consensus_node_ids)
                print(f"DEBUG: Consensus nodes: {consensus_nodes}")
            except Exception as e:
                print(f"DEBUG: Error getting consensus nodes: {e}")
                consensus_nodes = set()  # Fallback to empty set if consensus fails

            # Create nodes using the same logic as jsOutput
            for node in graph.nodeiterator()():
                title_text = ""
                if node.sequences:
                    title_text += f"Sequences: {node.sequences}"
                if node.variations:
                    title_text += ";;;".join(
                        [f"{sequence_id}: {text}" for sequence_id, text in node.variations.items()]
                    )
                    title_text = title_text.replace('"', "'")

                # Use the same color logic as jsOutput
                color = "#ceeab2" if node.ID in consensus_nodes else "#cae0e6"

                node_data = {
                    "id": node.ID,
                    "label": f"{node.ID}: {node.text}",
                    "title": title_text,
                    "color": color,
                }
                nodes.append(node_data)

            print(f"DEBUG: Created {len(nodes)} nodes")

            # Create edges using the same logic as jsOutput
            for node in graph.nodeiterator()():
                nodeID = node.ID  # Keep as integer
                for edge in node.outEdges:
                    target = edge  # Keep as integer
                    weight = node.outEdges[edge].weight + 1.5
                    edge_data = {
                        "from": nodeID,
                        "to": target,
                        "value": weight,
                        "color": "#cae0e6",
                        "arrows": "to",
                    }
                    edges.append(edge_data)

            print(f"DEBUG: Created {len(edges)} edges")
        except Exception as e:
            print(f"DEBUG: Error processing graph data: {e}")
            return jsonify({"error": f"Error processing graph data: {str(e)}"}), 500

        # Extract text from consensus nodes
        consensus_text = ""
        try:
            consensus_node_texts = []
            for node in graph.nodeiterator()():
                if node.ID in consensus_nodes and node.text and node.text.strip():
                    consensus_node_texts.append(node.text.strip())
            consensus_text = " ".join(consensus_node_texts)
        except Exception:
            consensus_text = ""

        # Check if we should compute consensus using decode_consensus
        compute_consensus = data.get("compute_consensus", False)
        if compute_consensus and decode_consensus:
            try:
                # Default to "bio" task for new graphs
                consensus_text = decode_consensus(graph, selection_threshold=0.5, task="bio")
            except Exception as e:
                print(f"DEBUG: Error computing consensus with decode_consensus: {e}")
                # Keep the original consensus text if decode_consensus fails

        # Get original sequences
        try:
            raw_sequences = graph._seqs if hasattr(graph, "_seqs") else []
            # Process sequences: join with spaces and remove "||"
            original_sequences = []
            for seq in raw_sequences:
                if isinstance(seq, list):
                    # Join list elements with spaces
                    processed_seq = " ".join(str(item) for item in seq)
                else:
                    processed_seq = str(seq)
                # Remove "||" characters
                processed_seq = processed_seq.replace("||", "")
                original_sequences.append(processed_seq)
        except Exception:
            original_sequences = []

        print("DEBUG: Returning success response")
        return jsonify(
            {
                "success": True,
                "nodes": nodes,
                "edges": edges,
                "num_sequences": len(sequences),
                "num_nodes": len(nodes),
                "num_edges": len(edges),
                "original_sequences": original_sequences,
                "consensus_text": consensus_text,
            }
        )

    except Exception as e:
        print(f"DEBUG: Main exception in create_graph: {e}")
        return jsonify({"error": str(e)}), 500


@app.route("/api/save_graph", methods=["POST"])
def save_graph():
    """Save a POA graph to a pickle file"""
    try:
        data = request.get_json()
        sequences = data.get("sequences", [])
        filename = data.get("filename", "graph.pkl")

        if len(sequences) < 2:
            return jsonify({"error": "At least 2 sequences are required"}), 400

        # Create the graph
        graph = TextPOAGraph(sequences[0], label=0)

        # Add remaining sequences
        for i, sequence in enumerate(sequences[1:], 1):
            alignment = TextSeqGraphAlignment(
                text=sequence,
                graph=graph,
                fastMethod=True,
                globalAlign=True,
                matchscore=1,
                mismatchscore=-2,
                gap_open=-1,
            )
            graph.incorporateSeqAlignment(alignment, sequence, label=i)

        # Refine the graph
        graph.refine_graph(verbose=False)

        # Save to pickle file
        graph.save_to_pickle(filename)

        return jsonify(
            {"success": True, "filename": filename, "message": f"Graph saved to {filename}"}
        )

    except Exception as e:
        return jsonify({"error": str(e)}), 500


@app.route("/api/graph_info", methods=["POST"])
def graph_info():
    """Get information about a graph without creating the full visualization"""
    try:
        data = request.get_json()
        sequences = data.get("sequences", [])

        if len(sequences) < 2:
            return jsonify({"error": "At least 2 sequences are required"}), 400

        # Create the graph
        graph = TextPOAGraph(sequences[0], label=0)

        # Add remaining sequences
        for i, sequence in enumerate(sequences[1:], 1):
            alignment = TextSeqGraphAlignment(
                text=sequence,
                graph=graph,
                fastMethod=True,
                globalAlign=True,
                matchscore=1,
                mismatchscore=-2,
                gap_open=-1,
            )
            graph.incorporateSeqAlignment(alignment, sequence, label=i)

        # Refine the graph
        graph.refine_graph(verbose=False)

        # Get consensus response
        consensus_text = graph.consensus_response()

        return jsonify(
            {
                "success": True,
                "num_sequences": len(sequences),
                "num_nodes": graph._nnodes,
                "consensus_text": consensus_text,
                "consensus_node_ids": graph.consensus_node_ids,
            }
        )

    except Exception as e:
        return jsonify({"error": str(e)}), 500


if __name__ == "__main__":
    print("Starting POA Graph Web Interface Server...")
    print("Open http://localhost:8080 in your browser")
    app.run(debug=True, host="0.0.0.0", port=8080)