567 points by ml_enthusiast 5 months ago flag hide 10 comments
user1 5 months ago next
Interesting project, I'd love to see more details about the data you're using and how you're training the model.
user2 5 months ago next
@user1 The data is a combination of financial statements and news articles, and I'm using a LSTM network with attention for training.
user3 5 months ago next
@user2 LSTM and attention, very cool! I've been looking into using LSTMs for time-series data, but haven't used attention yet.
user4 5 months ago prev next
What kind of accuracy are you getting on the predictions?
user5 5 months ago next
@user4 I'm currently getting around 75% accuracy on a 3-day rolling window, but I'm still tweaking the hyperparameters to improve it.
user6 5 months ago next
@user5 That's not bad, especially for unstructured data like financial news articles. Have you considered using a method like cross-validation to get a better estimate of the model's performance?
user7 5 months ago prev next
What kind of preprocessing are you doing on the news articles?
user8 5 months ago next
@user7 I'm using a combination of NLTK and spaCy for NLP tasks, such as tokenization, stopword removal, and lemmatization. I'm also using some custom heuristics to extract company-specific information from the articles.
user9 5 months ago next
@user8 That's a lot of work! Have you considered using a pre-trained language model for some of the NLP tasks? It might save you some time and effort.
user10 5 months ago prev next
This is a great project, keep up the good work!