import os import dgl import torch import pandas as pd import ast import numpy as np import joblib import json from transformers import AutoModelForSequenceClassification, AutoTokenizer from models import ModifiedGCN # Function to load and clean JSON data def load_and_clean_json_data(file): def clean_value(value): try: if '.' in value: return float(value) else: return int(value) except ValueError: return value with open(file, 'r') as f: data = json.load(f) for entry in data: for key, value in entry.items(): if isinstance(value, str) and value.replace('.', '', 1).replace('-', '', 1).isdigit(): entry[key] = clean_value(value) elif isinstance(value, str) and (value.startswith('[') and value.endswith(']')): entry[key] = ast.literal_eval(value) return data # Function to process the file and return status and processed file def process_file(file): if file is not None: status = "Processing" scaler_path = 'label_scaler.pkl' label_scaler = joblib.load(scaler_path) graphs = {} try: site_data = load_and_clean_json_data(file) except Exception as e: return f"Error loading JSON data: {e}", None combination_name = file.name.replace('.json', '') src_nodes = [] dst_nodes = [] node_features = [] edge_features = [] edge_weights = [] edge_set = set() for site in site_data: try: site['children_id'] = ast.literal_eval(str(site['children_id'])) site['angle_node_children'] = ast.literal_eval(str(site['angle_node_children'])) site['edge_length'] = ast.literal_eval(str(site['edge_length'])) except Exception as e: return f"Error processing site data: {e}", None node_features.append([ site['num_children'], site['x_coordinate'], site['y_coordinate'], site['z_coordinate'], site['area'], site['volume'], site['avg_normals_x'], site['avg_normals_y'], site['avg_normals_z'], site['bb_depth'], site['bb_width'], site['bb_height'], site['surface_area'], site['density'], site['vertices_faces_ratio'], site['vertices_edges_ratio'] ]) src_node = site['node_id'] children_ids = site['children_id'] angle_node_children = site['angle_node_children'] edge_length = site['edge_length'] for child_id, angle, length in zip(children_ids, angle_node_children, edge_length): edge = (src_node, child_id) reverse_edge = (child_id, src_node) if edge not in edge_set and reverse_edge not in edge_set: src_nodes.append(src_node) dst_nodes.append(child_id) edge_features.append([angle, length]) edge_weights.append(length) # Use edge length as weight edge_set.add(edge) edge_set.add(reverse_edge) # Ensure reverse edge is also added # Create graph g = dgl.graph((src_nodes, dst_nodes)) # Add node features node_features_tensor = torch.tensor(node_features, dtype=torch.float32) # Append meteorological data to each node's features meteorological_data = [ site_data[0]['prevailing_wind'], site_data[0]['avg_wind_m/s'], site_data[0]['yearly_overall_sun_h'], site_data[0]['yearly_overall_radiation_wh/sqm'], site_data[0]['yearly_avg_temp'], site_data[0]['min_temp'], site_data[0]['max_temp'] ] meteo_tensor = torch.tensor(meteorological_data, dtype=torch.float32).view(1, -1) meteo_tensor = meteo_tensor.repeat(g.num_nodes(), 1) # Repeat for each node combined_features_tensor = torch.cat((node_features_tensor, meteo_tensor), dim=1) # Concatenate feature-wise # Apply bipolar log transformation combined_features_log = torch.sign(combined_features_tensor) * torch.log1p(torch.abs(combined_features_tensor)) g.ndata['feature'] = combined_features_log # Add edge features g.edata['feature'] = torch.tensor(edge_features, dtype=torch.float32) # Add edge weights g.edata['weight'] = torch.tensor(edge_weights, dtype=torch.float32) graphs[combination_name] = g print(f'Processed graph for combination {combination_name}: {g.num_nodes()} nodes, {g.num_edges()} edges.') try: # Load the model with the custom class model_path = 'complete_model.pth' model = torch.load(model_path, map_location=torch.device('cpu')) except Exception as e: return f"Error loading model: {e}", None # Define new_graphs using graphs dictionary values new_graphs = [dgl.add_self_loop(g) for g in graphs.values()] new_batch = dgl.batch(new_graphs) new_features = new_batch.ndata['feature'] new_edge_features = new_batch.edata['feature'] new_edge_weights = new_batch.edata['weight'] model.eval() with torch.no_grad(): predictions = model(new_batch, new_features, new_edge_features, new_edge_weights) predictions = label_scaler.inverse_transform(predictions.numpy()) # Extract combination name combination_name = os.path.splitext(os.path.basename(file.name))[0] # Define column names for the predictions columns = [ 'avg_wind_around_combination_m/s', 'area_percentage_low_wind', 'avg_radiation_walls_wh/sqm', 'area_percentage_high_radiation_wh/sqm', 'overall_radiation_walls_wh/sqm', 'area_percentage_visibility>50%', 'egress_distance_sum_shortest_paths_m', 'egress_distance_num_paths>60m', 'yearly_energy_consumption_kWh' ] # Convert predictions to list of dicts and ensure all values are float predictions_list = [] for pred in predictions: pred_dict = {col: float(pred_val) for col, pred_val in zip(columns, pred)} predictions_list.append(pred_dict) # Create a dictionary with combination name and predictions data = { 'combination_name': combination_name, 'predictions': predictions_list } # Save the data to a JSON file output_file = 'predictions.json' with open(output_file, 'w') as f: json.dump(data, f, indent=4) status = "Finished" return status, output_file return "No file uploaded.", None