Anupam251272's picture
Create app.py
c44f410 verified
import os
import numpy as np
import torch
import matplotlib.pyplot as plt
import networkx as nx
import gradio as gr
from matplotlib.colors import LinearSegmentedColormap
import matplotlib.patches as mpatches
# Check if GPU is available
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")
class EnhancedMindMapGenerator:
def __init__(self):
self.graph = nx.DiGraph() # Using DiGraph for directed edges
self.node_positions = {}
self.node_colors = {}
self.edge_colors = {}
self.node_sizes = {}
self.node_depth = {}
self.levels = {}
def reset(self):
self.graph = nx.DiGraph()
self.node_positions = {}
self.node_colors = {}
self.edge_colors = {}
self.node_sizes = {}
self.node_depth = {}
self.levels = {}
return "Mind map reset successfully"
def parse_input(self, text):
"""Parse the input text into nodes and relationships"""
lines = text.strip().split('\n')
root_node = None
parent_map = {} # Track parent nodes based on indent level
current_indent_level = -1
current_parent = None
# First pass: Build hierarchy based on indentation
for line in lines:
original_line = line
line = line.strip()
if not line or '->' in line:
continue # Skip empty lines and relationship lines for now
# Calculate indent level
indent_level = len(original_line) - len(original_line.lstrip())
if root_node is None:
# This is the root node
root_node = line
self.add_node(root_node, is_root=True, depth=0)
parent_map[0] = root_node
current_indent_level = indent_level
current_parent = root_node
self.levels[0] = [root_node]
else:
# Handle indentation to determine parent-child relationships
if indent_level > current_indent_level:
# This is a child of the previous node
parent_map[indent_level] = current_parent
parent = None
if indent_level in parent_map:
parent = parent_map[indent_level]
# If this is a new indent level, set the parent to the previous node
if indent_level > current_indent_level:
parent = current_parent
else:
# Find the closest parent based on indent
closest_indent = max([i for i in parent_map.keys() if i < indent_level], default=0)
parent = parent_map[closest_indent]
# Calculate depth based on parent's depth
parent_depth = self.node_depth.get(parent, 0)
current_depth = parent_depth + 1
# Add node and edge
self.add_node(line, depth=current_depth)
self.add_edge(parent, line, "hierarchy")
# Add to level structure
if current_depth not in self.levels:
self.levels[current_depth] = []
self.levels[current_depth].append(line)
# Update tracking variables
current_indent_level = indent_level
current_parent = line
parent_map[indent_level] = line
# Second pass: Process explicit relationships (->)
for line in lines:
line = line.strip()
if '->' in line:
parts = line.split('->')
if len(parts) == 2:
source = parts[0].strip()
target = parts[1].strip()
self.add_edge(source, target, "relationship")
return f"Parsed mind map with root: {root_node}"
def add_node(self, node_name, is_root=False, depth=0):
"""Add a node to the graph"""
if node_name not in self.graph.nodes:
self.graph.add_node(node_name)
self.node_depth[node_name] = depth
# Set color based on depth
if is_root:
self.node_colors[node_name] = '#FF5733' # Root is red
self.node_sizes[node_name] = 2500
else:
# Use a color scheme based on depth
color_map = {
1: '#3498DB', # Blue
2: '#F39C12', # Orange
3: '#2ECC71', # Green
4: '#9B59B6', # Purple
5: '#E74C3C', # Red
}
self.node_colors[node_name] = color_map.get(depth % len(color_map), '#95A5A6') # Gray as default
self.node_sizes[node_name] = 2000 - (depth * 200) # Size decreases with depth
def add_edge(self, source, target, edge_type="hierarchy"):
"""Add an edge between two nodes"""
if source not in self.graph.nodes:
self.add_node(source)
if target not in self.graph.nodes:
self.add_node(target)
if not self.graph.has_edge(source, target):
self.graph.add_edge(source, target)
# Color edges based on type
if edge_type == "relationship":
self.edge_colors[(source, target)] = 'green'
else:
self.edge_colors[(source, target)] = 'gray'
def calculate_hierarchical_layout(self):
"""Calculate a hierarchical layout based on node depth"""
# Use hierarchical layout with depth levels
pos = {}
max_nodes_per_level = max([len(nodes) for nodes in self.levels.values()])
for level, nodes in self.levels.items():
y = -level * 2 # Vertical position based on level
# Center the nodes at each level
width = max(max_nodes_per_level, len(nodes))
for i, node in enumerate(nodes):
x = (i - (len(nodes) - 1) / 2) * 3 # Horizontal spacing
pos[node] = np.array([x, y])
return pos
def optimize_layout(self):
"""Use GPU-accelerated optimization for node layout (if available)"""
# First set initial positions using hierarchical layout
initial_pos = self.calculate_hierarchical_layout()
self.node_positions = initial_pos
if device.type == "cuda":
print("Optimizing layout using GPU...")
# Implement GPU optimization if needed
nodes = list(self.graph.nodes)
positions = torch.tensor([self.node_positions[node] for node in nodes], device=device)
# Simple force-directed algorithm using PyTorch (maintains hierarchical structure)
for _ in range(50):
# Calculate attractive forces (edges)
attractive_force = torch.zeros_like(positions)
for u, v in self.graph.edges:
u_idx = nodes.index(u)
v_idx = nodes.index(v)
direction = positions[v_idx] - positions[u_idx]
distance = torch.norm(direction) + 1e-5
force = direction * torch.log(distance / 2) * 0.1
attractive_force[u_idx] += force
attractive_force[v_idx] -= force
# Calculate repulsive forces (nodes at same level)
repulsive_force = torch.zeros_like(positions)
for level_nodes in self.levels.values():
level_indices = [nodes.index(node) for node in level_nodes if node in nodes]
for i_idx, i in enumerate(level_indices):
for j in level_indices[i_idx+1:]:
direction = positions[j] - positions[i]
distance = torch.norm(direction) + 1e-5
if distance < 3.0: # Only apply repulsion when nodes are close
force = direction / (distance ** 2) * 0.5
repulsive_force[i] -= force
repulsive_force[j] += force
# Update positions but maintain y-coordinate (level)
new_pos = positions + (attractive_force + repulsive_force) * 0.1
# Preserve y-coordinates to maintain hierarchical layout
for i, node in enumerate(nodes):
level = self.node_depth[node]
new_pos[i, 1] = positions[i, 1] # Keep original y-coordinate
positions = new_pos
# Copy back to CPU and update positions
positions_cpu = positions.cpu().numpy()
for i, node in enumerate(nodes):
self.node_positions[node] = positions_cpu[i]
return "Layout optimized using GPU acceleration while preserving hierarchy"
else:
# CPU-based optimization
# Adjust positions to prevent overlaps while maintaining hierarchy
pos = nx.spring_layout(
self.graph,
pos=self.node_positions,
fixed=None, # Don't fix positions
k=1.5, # Increase node separation
iterations=50,
weight=None
)
# Preserve y-coordinates to maintain hierarchical layout
for node in self.graph.nodes:
pos[node][1] = self.node_positions[node][1] # Keep original y-coordinate
self.node_positions = pos
return "Layout optimized using CPU while preserving hierarchy"
def visualize(self):
"""Generate a visualization of the mind map"""
if not self.graph.nodes:
return None
plt.figure(figsize=(16, 12), dpi=100)
# Use calculated positions from hierarchical layout or optimization
pos = self.node_positions
# Create a legend for depth levels
depth_colors = {}
for node, depth in self.node_depth.items():
if depth not in depth_colors:
depth_colors[depth] = self.node_colors[node]
# Draw edges with curved arrows for relationships
for edge in self.graph.edges:
edge_color = self.edge_colors.get(edge, 'gray')
# Use curved edges for explicit relationships, straight for hierarchy
if edge_color == 'green':
nx.draw_networkx_edges(
self.graph,
pos,
edgelist=[edge],
width=2.5,
edge_color=edge_color,
alpha=0.8,
arrows=True,
arrowsize=15,
connectionstyle="arc3,rad=0.3"
)
else:
nx.draw_networkx_edges(
self.graph,
pos,
edgelist=[edge],
width=1.5,
edge_color=edge_color,
alpha=0.7,
arrows=True,
arrowsize=12
)
# Draw nodes with depth-based colors
for node in self.graph.nodes:
nx.draw_networkx_nodes(
self.graph,
pos,
nodelist=[node],
node_color=self.node_colors.get(node, 'blue'),
node_size=self.node_sizes.get(node, 1000),
alpha=0.9,
edgecolors='black',
linewidths=1
)
# Draw labels with white background for better readability
label_pos = {node: (pos[node][0], pos[node][1]) for node in self.graph.nodes}
nx.draw_networkx_labels(
self.graph,
label_pos,
font_size=10,
font_family='sans-serif',
font_weight='bold',
bbox=dict(facecolor='white', alpha=0.7, edgecolor='none', boxstyle='round,pad=0.3')
)
# Add a legend
legend_elements = [
mpatches.Patch(color='#FF5733', label='Root'),
mpatches.Patch(color='#3498DB', label='Level 1'),
mpatches.Patch(color='#F39C12', label='Level 2'),
mpatches.Patch(color='#2ECC71', label='Level 3'),
mpatches.Patch(color='#9B59B6', label='Level 4+'),
mpatches.Patch(color='green', label='Explicit Relationship'),
mpatches.Patch(color='gray', label='Hierarchical Relationship')
]
plt.legend(handles=legend_elements, loc='upper right')
plt.title("Mind Map Visualization", fontsize=16, fontweight='bold')
plt.axis('off')
plt.tight_layout()
# Save to a temporary file
temp_path = "mindmap_output.png"
plt.savefig(temp_path, format="png", dpi=300, bbox_inches='tight', facecolor='white')
plt.close()
return temp_path
# Create the Gradio interface
def create_mind_map(input_text, optimization):
"""Create a mind map from input text"""
generator = EnhancedMindMapGenerator()
message = generator.parse_input(input_text)
print(message)
if optimization:
message = generator.optimize_layout()
print(message)
image_path = generator.visualize()
return image_path
# For Colab, use this function to create and launch the demo
def create_and_launch():
"""Create and launch the Gradio demo"""
with gr.Blocks(theme=gr.themes.Soft()) as demo:
gr.Markdown("# Enhanced Mind Map Generator")
gr.Markdown("Enter your mind map structure below. Use indentation to represent hierarchy or use -> for direct relationships.")
with gr.Row():
with gr.Column(scale=2):
input_text = gr.Textbox(
placeholder="Project Name\n Task 1\n Subtask 1.1\n Subtask 1.2\n Task 2\nTask 1 -> Task 2",
label="Mind Map Structure",
lines=15
)
with gr.Row():
optimization = gr.Checkbox(label="Use Layout Optimization", value=True)
generate_btn = gr.Button("Generate Mind Map", variant="primary")
gr.Markdown("### Format Guide:")
gr.Markdown("""
- Use indentation (spaces) to create parent-child relationships
- Each level of indentation creates a new depth level
- Use '-> ' to create explicit connections (e.g., 'NodeA -> NodeB')
- The first non-indented line becomes the root node
""")
with gr.Column(scale=3):
output_image = gr.Image(label="Generated Mind Map", type="filepath")
generate_btn.click(fn=create_mind_map, inputs=[input_text, optimization], outputs=output_image)
# Add examples
example_input1 = """Software Project
Planning
Requirements Gathering
Project Timeline
Resource Allocation
Development
Frontend
UI Design
React Components
Backend
API Development
Database Setup
Testing
Unit Tests
Integration Tests
Deployment
CI/CD Pipeline
Production Release
Planning -> Development
Development -> Testing
Testing -> Deployment"""
example_input2 = """Business Strategy
Market Analysis
Customer Demographics
Competitor Research
Market Trends
Internal Assessment
SWOT Analysis
Resource Inventory
Strategic Goals
Short-term Objectives
Long-term Vision
Implementation
Action Plans
Market Analysis -> Strategic Goals
Internal Assessment -> Strategic Goals
Strategic Goals -> Implementation"""
gr.Examples(
examples=[[example_input1, True], [example_input2, True]],
inputs=[input_text, optimization],
outputs=output_image,
fn=create_mind_map,
cache_examples=True,
)
# Launch with sharing enabled for Colab
demo.launch(share=True, debug=True)
return demo
# Main execution
def run_in_colab():
# Install necessary packages
print("Installing required packages...")
try:
import gradio
import networkx
except ImportError:
!pip install gradio networkx matplotlib
print("Packages installed!")
# Create and launch the demo
print("Launching the Enhanced Mind Map Generator...")
create_and_launch()
# For Google Colab, use this
try:
import google.colab
print("Running in Google Colab environment")
run_in_colab()
except:
print("Running in local environment")
# If not in Colab, just create and launch
create_and_launch()