1200 points by ai_engineer 6 months ago flag hide 16 comments
finance_engineer 6 months ago next
Interesting project! I'd love to hear more about the machine learning algorithms you used to predict stock prices. Have you considered testing it on a variety of different stocks and markets?
ml_enthusiast 6 months ago next
@finance_engineer I used a combination of LSTM and random forest algorithm for my project. I will definitely consider testing it on various stocks and markets for better accuracy. Thanks for the suggestion!
hugonottech 6 months ago prev next
Can you share more details about your data preprocessing? What kind of data did you use to train your model?
ml_enthusiast 6 months ago next
@hugonottech I used financial metrics like P/E ratio, EPS, dividend yield along with historical stock prices for data preprocessing. I also used web scraping to extract news articles related to the stock, incorporating NLP techniques to extract features.
datajunkie 6 months ago prev next
This is a great starting point, but what about taking into account external factors such as geopolitical events or changes in regulations?
ml_enthusiast 6 months ago next
@datajunkie I completely agree! I have plans to incorporate external factors into my model as well. Bringing in real-time data feeds and regulatory changes, will be my next steps.
statsmaven 6 months ago prev next
Have you compared your predictions with other existing models? How does it hold up?
ml_enthusiast 6 months ago next
@statsmaven I haven’t compared it against other models yet, but I plan to as part of my ongoing project. I will be sure to report back on my findings!
codewhisperer 6 months ago prev next
What kind of infrastructure did you use to train your model? Were you able to implement a distributed training system?
ml_enthusiast 6 months ago next
@codewhisperer I used a Google Colab VM to train my model, as I have access to a GPU. I haven't implemented distributed training system but plan to set up a distributed training infrastructure soon.
nlp_hacker 6 months ago prev next
How did you handle overfitting in your LSTM network? Did you use regularization techniques like dropout or recurrent dropout?
ml_enthusiast 6 months ago next
@nlp_hacker Yes, I used both dropout and recurrent dropout during my training process. I'll be sharing more information about addressing overfitting in my detailed write-up.
stockguru 6 months ago prev next
How do you plan on optimizing for latency while deploying to production? Training and prediction time might really matter for live trading signals.
ml_enthusiast 6 months ago next
@stockguru You raise a great point. I have looked into using TensorFlow Serving for model deployment and loading my models in memory. This combined with batch inference should help lower latency.
backtestboss 6 months ago prev next
Have you done any backtesting to assess the risk-adjusted returns of your strategy? It's crucial to know whether the model performs well without overfitting to historical data.
ml_enthusiast 6 months ago next
@backtestboss I realize using historical data to train the model is a common caveat. I've used walk-forward analysis in my backtesting strategy, taking into account partial data for a more robust test. I'll dive deeper into assessing risk-adjusted returns in the future!