this post was submitted on 17 Nov 2023
1 points (100.0% liked)
Machine Learning
1 readers
1 users here now
Community Rules:
- Be nice. No offensive behavior, insults or attacks: we encourage a diverse community in which members feel safe and have a voice.
- Make your post clear and comprehensive: posts that lack insight or effort will be removed. (ex: questions which are easily googled)
- Beginner or career related questions go elsewhere. This community is focused in discussion of research and new projects that advance the state-of-the-art.
- Limit self-promotion. Comments and posts should be first and foremost about topics of interest to ML observers and practitioners. Limited self-promotion is tolerated, but the sub is not here as merely a source for free advertisement. Such posts will be removed at the discretion of the mods.
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Looking forward to it, great job so far!
I have a question about applying test sets to models. Although I planned to email you, but i found you here by random!
My master's thesis in Finance was based on Easley et al.'s 2020 paper "Microstructure in the Machine Age" from the Review of Financial Studies. This paper used Random Forest (RF) to analyze financial market microstructures. The researchers created a RF-model with dataset of several microstructure indicator. Then evaluated which features were most significant in classifying the new unseen data (test-set). We do not look at prediction in this case, we look at the which features split the most effectively on the test-set.
In our tests, several indicators had high feature importance when creating the RF-model. But when applied to new, unseen data (the test set), the importance of these features often changed. Features that were crucial in the training phase were not as important for splitting new data, a pattern we consistently observed.
Easley et al. (2020) suggest that this occurs because financial microstructures are traditionally constructed on "in-sample" data, which has limited predictive power for new, unseen data.
I believe this finding can be important, as it can serve as an alternative way to back-testing. As backtesting does not make any sense for financial data, as all the random states in the market has already known.
I'm wondering if the Tsetlin Machine could be used to measure and analyze the same idea?
Sounds like an exciting problem! I guess leveraging the Tsetlin machine clauses can give a fresh take on the task. Tsetlin machines also support reasoning by elimination. That is, it can learn what the target isn’t instead of what it is, for increased robustness: https://www.ijcai.org/proceedings/2022/616