FractalAI / app.py
Aluode's picture
Upload 3 files
fda9246 verified
import numpy as np
import pickle
import gradio as gr
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from io import BytesIO
from PIL import Image
import random
import requests
from bs4 import BeautifulSoup
import time
import networkx as nx
# Constants
MAX_DEPTH = 15
MAX_CHILDREN = 5
SPACE_SIZE = 10
GROWTH_PROBABILITY = 0.2 # Increased from 0.1
class FractalNode:
def __init__(self, node_id, position):
self.id = node_id
self.position = position
self.connections = {}
self.activation = 0.0
def activate(self, input_signal):
self.activation = np.tanh(input_signal)
def connect(self, other_node, weight):
self.connections[other_node.id] = weight
class FractalNetwork:
def __init__(self, initial_nodes=5, space_size=SPACE_SIZE):
self.nodes = {}
self.space_size = space_size
self.graph = nx.Graph()
self.cycle_count = 0
self.memory = ""
self.create_initial_nodes(initial_nodes)
def create_initial_nodes(self, num_nodes):
for i in range(num_nodes):
position = np.random.rand(3) * self.space_size
self.add_node(FractalNode(i, position))
def add_node(self, node):
self.nodes[node.id] = node
self.graph.add_node(node.id, pos=node.position)
def connect_nodes(self, node1, node2, weight):
node1.connect(node2, weight)
node2.connect(node1, weight)
self.graph.add_edge(node1.id, node2.id, weight=weight)
def grow(self):
new_node_id = len(self.nodes)
position = np.random.rand(3) * self.space_size
new_node = FractalNode(new_node_id, position)
self.add_node(new_node)
for node in self.nodes.values():
if node.id != new_node_id:
distance = np.linalg.norm(np.array(new_node.position) - np.array(node.position))
if distance < self.space_size * 0.2:
weight = np.random.rand()
self.connect_nodes(new_node, node, weight)
def hebbian_learning(self):
for node in self.nodes.values():
for other_node_id, weight in list(node.connections.items()):
other_node = self.nodes[other_node_id]
delta_weight = 0.01 * node.activation * other_node.activation
new_weight = np.clip(weight + delta_weight, 0, 1) # Clip weight to [0, 1]
node.connections[other_node_id] = new_weight
other_node.connections[node.id] = new_weight
self.graph[node.id][other_node_id]['weight'] = new_weight
def process_input(self, input_text):
input_signal = sum(ord(c) for c in input_text) / len(input_text) / 128
for node in self.nodes.values():
node.activate(input_signal)
self.hebbian_learning()
if random.random() < GROWTH_PROBABILITY:
self.grow()
def think(self):
self.cycle_count += 1
for node in self.nodes.values():
node.activate(np.random.rand())
self.hebbian_learning()
if random.random() < GROWTH_PROBABILITY:
self.grow()
return f"Cycle {self.cycle_count}: {chr(int(np.mean([node.activation for node in self.nodes.values()]) * 26) + 97)}"
def chat(self, input_text):
self.memory += input_text + " "
if len(self.memory) > 1000:
self.memory = self.memory[-1000:]
self.process_input(input_text)
response = ''.join(random.choice(self.memory) for _ in range(20))
self.cycle_count += 1
return f"Cycle {self.cycle_count}: {response}"
def save_state(self, filename):
with open(filename, 'wb') as f:
pickle.dump(self, f)
@staticmethod
def load_state(filename):
with open(filename, 'rb') as f:
return pickle.load(f)
def visualize(self, zoom=1.0):
fig = plt.figure(figsize=(10, 8))
ax = fig.add_subplot(111, projection='3d')
pos = nx.get_node_attributes(self.graph, 'pos')
for edge in self.graph.edges():
start = pos[edge[0]]
end = pos[edge[1]]
weight = self.graph[edge[0]][edge[1]]['weight']
ax.plot([start[0], end[0]], [start[1], end[1]], [start[2], end[2]],
color='b', alpha=min(weight, 1.0), linewidth=weight*3)
for node_id, node_pos in pos.items():
ax.scatter(node_pos[0], node_pos[1], node_pos[2],
color='r', s=100*self.nodes[node_id].activation+50)
center = self.space_size / 2
ax.set_xlim(center - self.space_size/(2*zoom), center + self.space_size/(2*zoom))
ax.set_ylim(center - self.space_size/(2*zoom), center + self.space_size/(2*zoom))
ax.set_zlim(center - self.space_size/(2*zoom), center + self.space_size/(2*zoom))
plt.title(f"Fractal Network - {len(self.nodes)} nodes")
buf = BytesIO()
plt.savefig(buf, format='png')
buf.seek(0)
plt.close(fig)
image = Image.open(buf)
return image
def fetch_wikipedia_content(topic):
url = f"https://en.wikipedia.org/wiki/{topic}"
response = requests.get(url)
if response.status_code == 200:
soup = BeautifulSoup(response.content, 'html.parser')
paragraphs = soup.find_all('p')
content = ' '.join([p.text for p in paragraphs])
return content
else:
return None
def gradio_interface():
network = FractalNetwork()
zoom_level = 1.0
def cycle_ai(num_cycles):
nonlocal zoom_level
thoughts = []
for _ in range(num_cycles):
thought = network.think()
thoughts.append(thought)
image = network.visualize(zoom_level)
return "\n".join(thoughts), image
def save_state(filename):
if filename.strip() == "":
return "Please enter a valid filename."
try:
network.save_state(filename)
return f"Network state saved as {filename}"
except Exception as e:
return f"Error saving network state: {str(e)}"
def load_state(file):
if file is None:
return "Please upload a file."
try:
loaded_network = FractalNetwork.load_state(file.name)
nonlocal network
network = loaded_network
return f"Loaded network state from {file.name}"
except Exception as e:
return f"Error loading network state: {str(e)}"
def recreate_network(initial_nodes):
nonlocal network, zoom_level
network = FractalNetwork(initial_nodes=initial_nodes)
image = network.visualize(zoom_level)
return f"Network recreated with {initial_nodes} initial nodes", image
def train_on_wikipedia(topic):
nonlocal zoom_level
content = fetch_wikipedia_content(topic)
if content:
chunks = [content[i:i+500] for i in range(0, len(content), 500)]
thoughts = []
for chunk in chunks:
network.process_input(chunk)
thoughts.append(f"Processed chunk: {network.think()}")
image = network.visualize(zoom_level)
return "\n".join(thoughts), image
else:
return f"Could not retrieve content for topic: {topic}", None
def chat_with_ai(input_text):
nonlocal zoom_level
response = network.chat(input_text)
image = network.visualize(zoom_level)
return response, image
def self_conversation(num_cycles):
nonlocal zoom_level
thoughts = []
for _ in range(num_cycles):
thought = network.think()
thoughts.append(thought)
image = network.visualize(zoom_level)
time.sleep(0.1) # Add a small delay to make the process visible
yield "\n".join(thoughts), image
def update_zoom(zoom_factor):
nonlocal zoom_level
zoom_level *= zoom_factor
image = network.visualize(zoom_level)
return image
with gr.Blocks() as demo:
gr.Markdown("# Advanced Fractal AI with Visualization and Interaction")
with gr.Row():
num_cycles = gr.Number(label="Number of Cycles", value=1, precision=0)
cycle_button = gr.Button("Run Cycles")
output_text = gr.Textbox(label="AI Thoughts", lines=5)
fractal_viz = gr.Image(label="Fractal Visualization")
with gr.Row():
zoom_in = gr.Button("Zoom In")
zoom_out = gr.Button("Zoom Out")
with gr.Row():
save_name = gr.Textbox(label="Save filename:")
save_btn = gr.Button("Save Network State")
load_file = gr.File(label="Load Network State")
initial_nodes_slider = gr.Slider(minimum=1, maximum=20, step=1, value=5, label="Initial Nodes")
recreate_btn = gr.Button("Recreate Network")
wiki_topic = gr.Textbox(label="Wikipedia Topic:")
wiki_btn = gr.Button("Train on Wikipedia")
chat_input = gr.Textbox(label="Chat with Fractal AI")
chat_output = gr.Textbox(label="Fractal AI Response", lines=3)
chat_button = gr.Button("Send")
self_convo_cycles = gr.Number(label="Self-Conversation Cycles", value=10, precision=0)
self_convo_button = gr.Button("Start Self-Conversation")
# Connect components
cycle_button.click(cycle_ai, inputs=[num_cycles], outputs=[output_text, fractal_viz])
save_btn.click(save_state, inputs=[save_name], outputs=[output_text])
load_file.change(load_state, inputs=[load_file], outputs=[output_text])
recreate_btn.click(recreate_network, inputs=[initial_nodes_slider], outputs=[output_text, fractal_viz])
wiki_btn.click(train_on_wikipedia, inputs=[wiki_topic], outputs=[output_text, fractal_viz])
chat_button.click(chat_with_ai, inputs=[chat_input], outputs=[chat_output, fractal_viz])
self_convo_button.click(self_conversation, inputs=[self_convo_cycles], outputs=[output_text, fractal_viz])
zoom_in.click(update_zoom, inputs=[gr.State(1.2)], outputs=[fractal_viz])
zoom_out.click(update_zoom, inputs=[gr.State(0.8)], outputs=[fractal_viz])
return demo
# Launch the Gradio interface
if __name__ == "__main__":
demo = gradio_interface()
demo.launch()