Hyperparameter tuning neural networks on financial data

[brianrisk]

Fellow neural network enthusiasts

What’s the difference between ReLU and Leaky ReLU? Between binary cross-entropy, Huber loss and Poisson NLL? Between a learning rate of 0.01 and 0.001?

Does it really matter so long as you pick the right ones for your model?

I’m excited to put up this Python code aimed at simplifying this process of iterating through your desired hyperparameters. Modify the config.py file and the system manages the comprehensive search through potential configurations—either exhaustively via grid search or more selectively through random search; its multithreading functionality reduces compute time.

The repository includes sample stock data and optimizes towards precision p-value, critical for investing.

Code on GitHub

 

[Paul Lopez]

How has it worked out for you so far?

 

[Daniel Duffy]

Blogs - MSc Theses on Machine Learning and Computational Finance 2020 :: Datasim

MSc Thesis “The Performance of Artificial Neural Networks on Rough Heston Model” By Chun Kiat Ong, University of Birmingham U.K. 2020 Supervisor: Dr. Daniel J...

www.datasim.nl

Blogs - MSc Theses on Machine Learning and Computational Finance 2019 :: Datasim

MSc Theses at University of Birmingham 2019 MSc Mathematical Finance Programme Supervisor: Dr. Daniel J. Duffy, dduffy@datasim.nl Course Director: Dr. Colin Rowat,..

www.datasim.nl

 

[Paul Lopez]

Lovely papers sir!

 

[Paul Lopez]

I'd be curious if we have an implementation somewhere from scratch in C or C++.

The matrix multiplication stuff seems to be where things get interesting!

 

[Daniel Duffy]

I'd be curious if we have an implementation somewhere from scratch in C or C++.

The matrix multiplication stuff seems to be where things get interesting!

M1*M2 is kids' stuff.