Sigmoid activation function in option-pricing with R

tidy-finance / website

This repository hosts the source code for the website tidy-finance.org

https://tidy-finance.org

Other

82 stars 47 forks source link

Sigmoid activation function in option-pricing with R #104

Closed christophscheuch closed 4 months ago

christophscheuch commented 4 months ago

Updated brulee & parsnip
Changed activation functions
Added facet wrap to figure for alignment with TFWP
Fixed figure label

Open:

Need to set Python version to 3.10.11 (I forgot why it is 3.10.13 for website repo on my Mac) -> will do it after TFWP migration
Adjust the text of the results at the end

voigtstefan commented 4 months ago

Looking at the results, I wonder if we should update the entire section and include hyperparameter tuning just as we did in the chapter before. Currently, we fix all values, and as a result, deep NN seems completely off. These prediction errors should not be taken too serious as we did not bother to choose proper values but from a pedagogical point of view it would be the right thing to add.

christophscheuch commented 4 months ago

Maybe you looked at the wrong figure, there is a new one with ok-ish deep NN performance once I updated the penalty parameter: https://raw.githubusercontent.com/tidy-finance/website/28e39d86d80de5ba14aefc311705b9f2708ed577/_freeze/r/option-pricing-via-machine-learning/figure-html/fig-431-1.png

Let's discuss whether we want to include hyperparameter tuning in this chapter in a separate PR.