【粵語】AI炒股建模(EP1/2) | 真正有效的股價走勢預測AI，Python程式碼全公開，馬上跟著做吧！

Summary

In this video, the presenter introduces a powerful AI model designed to predict stock price trends. The video walks viewers through the process of developing the AI, from defining goals to building the dataset and model. It explains using a 100-day candlestick chart to predict daily stock movements, with a focus on achieving a 60% accuracy rate. The presenter discusses the technical details, including the choice of Tencent stock, and demonstrates the development platform setup. The video emphasizes practical coding demonstrations and scientific validation of the AI's effectiveness in stock market trading.

Takeaways

• 🧠 The speaker claims a scientific approach, emphasizing truth and transparency, and asserts that their stock prediction AI is effective with a statistical basis.
• 📈 The video discusses using AI to predict stock prices based on historical data, acknowledging the weak correlation between past and future prices but suggesting that accurate predictions can still lead to long-term profit.
• 🤖 The speaker had initially developed a stock-predicting AI that failed to meet expectations but has since improved the model using new techniques.
• 🧩 The AI model the speaker ultimately developed is named 'InceptionTime,' which is presented as more suitable for predicting stock price trends compared to simpler models.
• 🔍 The AI model's input data consists of 100-day candlestick charts (open, high, low, close prices) and volume, with the goal of predicting the next day's price movement.
• 📊 The AI is framed as a binary classifier, predicting whether the next day will be a bull market (up) or bear market (down), aiming for a 60% accuracy rate.
• 📅 The speaker uses daily candlestick charts to focus on making more frequent trades, while avoiding smaller timeframes like hours or minutes, as they are considered too noisy.
• 📊 The AI model is trained with historical stock data from Tencent Holdings (0700.HK), chosen for its 10-year history, significant price volatility, and large trading volume.
• 🛠️ The speaker demonstrates how to set up a development environment using TensorFlow and Docker, guiding viewers through the process to ensure compatibility across different platforms.
• 💻 The speaker walks through data preparation steps, creating datasets (training, validation, testing) using moving windows of 100 days, and balances the dataset for fair evaluation of bull and bear market predictions.

Q & A

What is the primary goal of the AI model discussed in the video?

The primary goal of the AI model is to predict whether a stock's price will rise (bull market) or fall (bear market) based on the past 100 days of price data, aiming for a prediction accuracy of over 60%.

Why is predicting stock prices based on past data challenging?

Predicting stock prices based on past data is challenging because there isn't a strong correlation between historical prices and future movements. However, the correlation is not zero, meaning AI might make useful predictions, but with limited accuracy.

Why does the video focus on daily stock price predictions rather than weekly or monthly?

The video focuses on daily predictions because more frequent trades provide more opportunities to accumulate small wins over time, leading to overall profits. Predicting on a smaller timescale like hourly or minute-based predictions is avoided due to the increased noise and difficulty in analyzing such short-term price movements.

What kind of data is used as input for the AI model?

The AI model uses daily candlestick charts (K charts), specifically the Open, High, Low, Close prices, and the trading volume for the past 100 days as input data.

How does the AI model classify the stock price trends?

The AI model is a binary classifier that outputs the probability of a bull market or a bear market. If the probability of a bull market is greater than 50%, the model predicts a price increase, otherwise, it predicts a price decrease.

What is the significance of maintaining a balance between bull and bear cases in the dataset?

A balanced dataset is essential for fair accuracy testing. Without balancing, a model could easily predict only bull markets during a rising trend and achieve a falsely high accuracy, but this wouldn't be useful in practice.

What is the role of the training, validation, and test datasets in the AI development process?

The training dataset is used to train the AI model, the validation dataset is used to prevent overfitting during training, and the test dataset is used to evaluate the model's performance on unseen data, ensuring the model can predict future stock movements accurately.

Why did the creator choose Tencent Holdings (0700.HK) as the stock for the AI model?

Tencent Holdings was chosen because it has over 10 years of historical price data, exhibits frequent price fluctuations (which is ideal for trading), and has a large trading volume, reducing the impact of market manipulation.

How does the model use the 100-day window of data to make predictions?

The model slides a 100-day window across the stock price history, using the data in each window to predict whether the stock price will rise or fall on the 101st day. This process is repeated to create a large dataset for training the AI.

What is InceptionTime, and why is it suitable for stock price prediction?

InceptionTime is a more advanced AI model that is effective at handling time-series data. It was chosen after the simpler Multilayer Perceptron model, as it showed better results in capturing the nuances of stock price movements and making more accurate predictions.