app.py

import os
import gradio as gr
import tensorflow as tf
import numpy as np
import cv2
from PIL import Image
import logging
from huggingface_hub import hf_hub_download
from huggingface_hub import login
import matplotlib.pyplot as plt
import matplotlib
matplotlib.use('Agg')  

# Настройка логирования
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Проверка наличия токена
if "HUGGINGFACE_TOKEN" not in os.environ:
    logger.error("HUGGINGFACE_TOKEN not found in environment variables!")
else:
    logger.info("HUGGINGFACE_TOKEN found")
    # Аутентификация с использованием токена
    login(token=os.environ["HUGGINGFACE_TOKEN"])
    logger.info("Logged in to Hugging Face")

# Определение размера изображения
IMG_SHAPE = (479, 1221, 3)

class SecureModel:
    _instance = None
    
    def __init__(self):
        try:
            logger.info("Attempting to download model files...")
            
            # Загружаем файл модели
            model_path = hf_hub_download(
                repo_id="Dianor/trading-model-private",
                filename="trading_modelbeta0.7.keras",
                token=os.environ["HUGGINGFACE_TOKEN"]
            )
            
            # Загружаем файл с кастомными слоями
            layers_path = hf_hub_download(
                repo_id="Dianor/trading-model-private",
                filename="custom_trading_layers.py",
                token=os.environ["HUGGINGFACE_TOKEN"]
            )
            
            logger.info(f"Files downloaded successfully")
            
            # Импортируем кастомные слои из скачанного модуля
            import importlib.util
            import sys
            
            # Загружаем модуль с кастомными слоями
            spec = importlib.util.spec_from_file_location("custom_trading_layers", layers_path)
            custom_module = importlib.util.module_from_spec(spec)
            sys.modules["custom_trading_layers"] = custom_module
            spec.loader.exec_module(custom_module)
            
            # Получаем словарь custom_objects
            custom_objects = custom_module.get_custom_objects()
            
            # Обновляем глобальные объекты TensorFlow
            tf.keras.utils.get_custom_objects().update(custom_objects)
            
            # Загружаем модель
            try:
                self.model = tf.keras.models.load_model(
                    model_path,
                    custom_objects=custom_objects,
                    compile=False
                )
            except Exception as load_error:
                logger.warning(f"Direct load failed: {load_error}")
                self.model = tf.keras.models.load_model(
                    model_path,
                    custom_objects=custom_objects,
                    compile=False
                )
                self.model.compile(
                    optimizer='adam',
                    loss={
                        'long_signal': 'binary_crossentropy',
                        'short_signal': 'binary_crossentropy'
                    },
                    metrics=['accuracy']
                )
                
            logger.info("Model loaded successfully")
        except Exception as e:
            logger.error(f"Failed to load model: {str(e)}")
            raise

    @classmethod
    def get_instance(cls):
        if cls._instance is None:
            cls._instance = cls()
        return cls._instance.model

def preprocess_image(image):
    try:
        logger.info(f"Starting preprocessing. Input shape: {image.shape}")
        
        # Конвертируем в RGB если нужно (если изображение в BGR)
        if len(image.shape) == 3 and image.shape[2] == 3:
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
            
        image = cv2.resize(image, (IMG_SHAPE[1], IMG_SHAPE[0]))  
        
        # Используем ту же нормализацию
        image = image.astype('float32') / 255.0
        
        logger.info(f"Preprocessed shape: {image.shape}")
        logger.info(f"Value range: [{image.min():.3f}, {image.max():.3f}]")
        
        return image
        
    except Exception as e:
        logger.error(f"Error in preprocess_image: {str(e)}")
        raise

def analyze_trading_chart(input_image):
    try:
        model = SecureModel.get_instance()
        
        # Сохраняем оригинал для отображения
        display_image = input_image.copy()
        
        # Логируем исходное изображение
        logger.info(f"Raw input shape: {input_image.shape}")
        logger.info(f"Raw input range: [{input_image.min()}, {input_image.max()}]")
        
        # Изменяем размер изображения до требуемого 
        resized_image = cv2.resize(input_image, (1221, 479), interpolation=cv2.INTER_NEAREST)
        
        # Нормализуем изображение
        image = resized_image.astype('float32') / 255.0
        image = np.expand_dims(image, axis=0)  # Добавляем batch dimension
        
        # Логируем после обработки
        logger.info(f"Processed input shape: {image.shape}")
        logger.info(f"Processed input range: [{image.min():.3f}, {image.max():.3f}]")
        
        # Делаем предсказание
        predictions = model.predict(image, verbose=0)
        
        # Логируем сырые предсказания
        logger.info(f"Raw predictions: {predictions}")
        
        long_signal = float(predictions['long_signal'][0][0])
        short_signal = float(predictions['short_signal'][0][0])
        
        logger.info(f"Final predictions: LONG={long_signal:.3f}, SHORT={short_signal:.3f}")
        
        # Создаем визуализацию
        plt.style.use('dark_background')
        fig = plt.figure(figsize=(15, 10), facecolor='#1E222D')
        gs = fig.add_gridspec(2, 1, height_ratios=[3, 1], hspace=0.3)
        
        # График цены
        ax1 = fig.add_subplot(gs[0])
        ax1.imshow(display_image)  # Показываем оригинальное изображение
        ax1.set_title('Trading Chart Analysis', color='#B7BDD7', pad=10, fontsize=14)
        ax1.axis('off')
        
        # Панель сигналов
        ax2 = fig.add_subplot(gs[1])
        ax2.set_facecolor('#1E222D')
        
        bar_positions = [0, 1]
        signal_values = [long_signal, short_signal]
        colors = ['#26a69a', '#ef5350']
        labels = ['Long Signal', 'Short Signal']
        
        bars = ax2.bar(bar_positions, signal_values, color=colors)
        ax2.set_xticks(bar_positions)
        ax2.set_xticklabels(labels, color='#B7BDD7', fontsize=12)
        ax2.set_ylim(0, 1)
        ax2.set_ylabel('Signal Strength', color='#B7BDD7', fontsize=12)
        ax2.grid(True, alpha=0.2)
        ax2.tick_params(colors='#B7BDD7')
        
        # Добавляем значения над барами
        for bar in bars:
            height = bar.get_height()
            ax2.text(bar.get_x() + bar.get_width()/2., height,
                    f'{height:.3f}',
                    ha='center', va='bottom', color='#B7BDD7',
                    fontsize=12)
        
        ax2.axhline(y=0.8, color='white', linestyle='--', alpha=0.5, label='Signal Threshold')
        ax2.legend(loc='upper right', bbox_to_anchor=(0.98, 0.98))
        
        # Конвертируем график в изображение
        fig.canvas.draw()
        buf = fig.canvas.buffer_rgba()
        img = np.asarray(buf)
        plt.close(fig)
        
        return img
        
    except Exception as e:
        logger.error(f"Error in analyze_trading_chart: {str(e)}")
        logger.exception("Full traceback:")
        return display_image

# Создаем интерфейс с табами
def create_interface():
    with gr.Blocks(theme=gr.themes.Default()) as demo:
        gr.Markdown("""
        # 🚀 Revolutionary Neural Vision Trading

        ## Next-Generation AI Computer Vision for Cryptocurrency Trading

        Introducing the world's first neural network that trades cryptocurrency through pure visual comprehension—a breakthrough technology that sees charts just like professional traders do.

        This revolutionary AI doesn't rely on traditional indicators or mathematical patterns. Instead, it employs advanced computer vision to interpret market dynamics visually, analyzing real-time price action with human-like perception but machine-level precision.

        The system provides confidence-based entry signals, automatically executing trades when conviction reaches 0.9 or higher—mimicking the decision-making process of elite traders while eliminating emotional bias.

        ---

        ### Unprecedented Market Understanding:

        ❗️ Visual price action analysis based on pure Computer Vision

        ❗️ Dynamic support/resistance identification through visual context

        ❗️ Real-time decision making focused on the critical last candle

        ---

        Try it now! Upload your TradingView/Binance dark theme chart, or use our examples and experience trading intelligence that exists nowhere else in the market.

        **💼 Limited partnership opportunities available for qualified investors. Contact us to join the visual trading revolution.**
        """)
        
        with gr.Tabs():
            # Таб анализа графиков
            with gr.Tab("Signal Analysis"):
                with gr.Row():
                    # Левая колонка для ввода
                    with gr.Column(scale=1):
                        gr.Markdown("""
                        ### Upload Your Trading Chart
                        Or use example charts below
                        """)
                        input_image = gr.Image(type="numpy", height=400)
                    
                    # Правая колонка для вывода
                    with gr.Column(scale=1):
                        gr.Markdown("""
                        ### Analysis Results
                        - Long Signal (Green): Upward movement probability
                        - Short Signal (Red): Downward movement probability
                        """)
                        output_image = gr.Image(type="numpy", height=400)
                
                analyze_btn = gr.Button("Analyze Chart", size="lg")
                analyze_btn.click(
                    fn=analyze_trading_chart,
                    inputs=input_image,
                    outputs=output_image
                )
                
                gr.Markdown("### Example Charts")
                gr.Examples(
                    examples=[
                        "example1.png", "example2.png", "example3.png", "example4.png", 
                        "example5.png", "example6.png", "example7.png", "example8.png", 
                        "example9.png", "example10.png"
                    ],
                    inputs=input_image,
                    outputs=output_image,
                    fn=analyze_trading_chart,
                    cache_examples=True,
                    examples_per_page=10  
                )
            
            # Таб с демонстрационным видео
            with gr.Tab("Trading Demo"):
                gr.Markdown("""
                ## Trading System Backtesting Demo
                Watch how our AI trading system performs in different market conditions.
                """)
                
                with gr.Row():
                    with gr.Column(scale=1, min_width=800):
                        gr.Video("demo.mp4")
                
                gr.Markdown("""
                ### What you're seeing in the demo:
                - Real-time trading decisions
                - Signal generation and execution
                - Performance metrics and profit visualization
                - Risk management in action
                """)

    return demo

if __name__ == "__main__":
    try:
        logger.info("Initializing model at startup...")
        SecureModel.get_instance()
        logger.info("Model initialized successfully")
    except Exception as e:
        logger.error(f"Failed to initialize model at startup: {str(e)}")
    
    demo = create_interface()
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False
    )