import gradio as gr
import numpy as np
import networkx as nx
import matplotlib.pyplot as plt
import pandas as pd
from sklearn.metrics import log_loss

def sigmoid(z):
    return 1.0/(1.0+ np.exp(-z))

def sigmoid_prime(z):
  return sigmoid(z)*(1.0-sigmoid(z))

def cross_entropy(weights,bias,X_train,y_train):
  return log_loss(y_train,[feedforward(X_train[k],bias,weights).reshape(-1,) for k in range(len(X_train))],labels=[1.0,0.0])

def feedforward(x,bias,weights):
    for b,w in zip(bias,weights):
      x=sigmoid(np.dot(w,x)+b)
    return x


def update_mini_batch(weights,bias,X_mini,y_mini,eta):
    derivative_b=[np.zeros(b.shape) for b in bias]
    derivative_w=[np.zeros(w.shape) for w in weights]
    for i in range(len(y_mini)):
      x_mi=X_mini[i]
      y_mi=y_mini[i]
      derivative_b_sample, derivative_w_sample=backprop(weights,bias,x_mi,y_mi)
      for j in range(len(derivative_b)):
        derivative_b[j]+=derivative_b_sample[j]
        derivative_w[j]+=derivative_w_sample[j]
    for i in range(len(weights)):
        weights[i]-=(eta/len(y_mini))*derivative_w[i]
        bias[i]-=(eta/len(y_mini))*derivative_b[i]
    return (weights,bias)

input_neurons = 10
hidden_neurons = 2
output_neurons = 1
weights_input_hidden = np.array([[ -7.22321659, -11.15471068, -30.00484398,  13.08466657,
        -15.0766471 ,  14.81645208, -13.11446119,  29.95009006,
          10.9668077 ,   7.11140227],
        [  7.48857155,  11.20225432,  30.07325613, -13.1441687 ,
          15.08768163, -15.10126949,  13.11893102, -30.1417512 ,
        -11.18918466,  -7.44585737]]).T
biases_hidden = np.array([[1.62426502],
        [1.63230885]]).reshape(-1)
weights_hidden_output = np.array([[23.73266741, 23.84381938]]).T
bias_output = np.array([[-35.80629779]]).reshape(-1)

def visualize_neural(a):
  buffer = io.BytesIO()
  input_values = [int(i) for i in list(a)]
  hidden_layer_input = np.dot(input_values, weights_input_hidden) + biases_hidden
  hidden_layer_output = np.array([round(i,3) for i in sigmoid(hidden_layer_input)])
  output_layer_input = np.dot(hidden_layer_output, weights_hidden_output) + bias_output
  output = sigmoid(output_layer_input)
  G = nx.DiGraph()
  for i in range(input_neurons):
      G.add_node(f'{i}={input_values[i]}', pos=(0, i))
  for i in range(hidden_neurons):
      G.add_node(f'{i},f({np.round(hidden_layer_input[i],3)})={np.round(hidden_layer_output[i],3)}', pos=(1, 0.5*(input_neurons-1-hidden_neurons)+i))
  for i in range(output_neurons):
      G.add_node(f'Output {i+1}={output}', pos=(2, 4))
  for i in range(input_neurons):
      for j in range(hidden_neurons):
          weight = weights_input_hidden[i][j]
          G.add_edge(f'{i}={input_values[i]}', f'{j},f({np.round(hidden_layer_input[j],3)})={np.round(hidden_layer_output[j],3)}', weight=f"W[{j,i}]={weight}")
  for i in range(hidden_neurons):
      for j in range(output_neurons):
          weight = weights_hidden_output[i][j]
          G.add_edge(f'{i},f({np.round(hidden_layer_input[i],3)})={np.round(hidden_layer_output[i],3)}', f'Output {j+1}={output}', weight=weight)
  pos = nx.get_node_attributes(G, 'pos')
  plt.figure(figsize=(10, 5),dpi=200)
  nx.draw(G, pos, with_labels=True, node_size=1e+3, node_color='lightblue', font_size=3, font_weight='bold')
  labels = nx.get_edge_attributes(G, 'weight')
  nx.draw_networkx_edge_labels(G, pos, edge_labels=labels,font_size=3,label_pos=0.8)
  plt.title('Neural Network Graph')
  plt.savefig(buffer, format='png')
  plt.close()
  buffer.seek(0)
  image = Image.open(buffer)
  image_array = np.array(image)
  if output<0.5:
      return image_array,"Non palindrom"
  else:
      return image_array,"palindrom"



demo = gr.Interface(fn=visualize_neural, inputs="text", outputs=[gr.Image(label="Neural network visualization"),gr.Textbox(label="output")])

demo.launch()