98 points by quant_guru 5 months ago flag hide 12 comments
financebot 5 months ago next
Fascinating project! Predicting stock prices with AI is a hot topic these days. Would love to hear more about the specific algorithms and techniques used.
codingfan 5 months ago next
From my understanding, the model utilizes LSTM (Long Short-Term Memory) recurrent neural networks combined with technical indicators to predict stock prices. However, the post could give more insights into tuning and optimizing the model.
codingfan 5 months ago next
@financebot: I re-read the post, and it seems that they fine-tuned the model with a combination of ADADELTA and ADAM optimizers. However, further details would indeed help understand the project better.
aiexpert 5 months ago prev next
@stats_guru: The author mentioned they used the mean squared error and mean absolute percentage error for evaluating the model, which are quite popular metrics for stock price predictions.
stats_guru 5 months ago prev next
What kind of evaluation metrics have you used to assess the model's performance? Cross validation for accuracy, precision and recall might give us more insight about this project.
modelmaster 5 months ago next
@stats_guru: There is actually a critique of using MSE as the main evaluation metric in the Hacker News comments back in tale 2531329, which you might want to read too.
devops_joe 5 months ago prev next
It would be interesting to learn more about backtesting results! Any open-source tools or packages used to conduct them?
quantsarah 5 months ago prev next
What kind of data preprocessing strategies did the author follow? I'd be curious to know more about the steps taken to avoid overfitting the model.
finance_fred 5 months ago prev next
What time horizon does the AI predict? Daily, Weekly, Monthly, etc? Also, what exchanges or markets are you covering? Ranging from NYSE to cryptocurrency markets would make quite a difference in the overall product.
geekgrl 5 months ago prev next
As a beginner in this field, I believe a tutorial on this subject would be incredibly helpful to those motivated by this project! Any chance the author would consider creating a guide or tutorial?