ML

Machine learning

The methods that work on tabular financial data, and why cross-validation is different here.

1
Framework
Bias–variance, overfitting, capacity and the curse of dimensionality.
- 1.1The framework: bias, variance, capacity and dimensionality12 min
2
Tree-based methods
Bagging, random forests and gradient boosting.
- 2.1Trees: bagging, random forests and gradient boosting12 min
3
Other supervised methods
kNN, SVMs, the kernel trick and generative versus discriminative.
- 3.1Other supervised methods: kNN, SVMs and the kernel trick11 min
4
Unsupervised learning
Clustering, GMM and EM, and correlation-matrix clustering.
- 4.1Unsupervised learning: clustering assets and correlation structure12 min
5
Model selection and evaluation
Why k-fold is wrong for time series: purging and embargoing.
- 5.1Why k-fold cross-validation is wrong on financial data12 min
6
Optimisation for learning
SGD, momentum, Adam and the EM algorithm.
- 6.1Optimisation: gradient descent, momentum and Adam11 min
7
Neural networks
Backpropagation, regularisation, and when deep learning is the wrong tool.
- 7.1Neural networks: backpropagation, and when they are the wrong tool12 min