Work in progress toolbox to estimate parameters of distributions in PyTorch to utilize GPU parallelization and automatic differentiation.
Incorporated algorithms:
- (MLE) Maximum Likelihood Estimation using gradient descent optimization
- (MAP) Maximum a Posteriori Estimation using gradient descent optimization
- (MM-MLE) Mixture Model Maximum Likelihood Estimation using gradient descent optimization
Given observations from a bimodal normal distribution with unknown parameters,
of which we want to estimate a target distribution (green in the figure),
utilizing a prior can greatly improve the estimator as seen in the figure.
The code for this plot can be found in the MAP implementation
and can be run via python -m stats.estimation.map.
The code for this plot can be found in the MM implementation
and can be run via python -m stats.estimation.mm.
