These are the outcomes of my learning during my internship (Jun-Aug 2020) in NAAMII. I wrote couple of scripts for finite and infinite mixture models based on Chinese Restaurant Process. I have shown usages of various conjugate priors and also collapsed sampling methods.
I worked along with Sushil Awale to implement unsupervised monolingual word segmentation based on goldwater-etal-2006-contextual and snyder-barzilay-2008-unsupervised. We came up with monolingual word segmentation with beta-geometric conjugate prior over the length of a given word.
- Finite Gaussian Mixture Model
- Infinite Gaussian Mixture Model
- Categorial/Multinomial Mixture Model with Dirichlet prior using MLE
- Categorial/Multinomial Mixture Model with Dirichlet prior using collapsed sampling
- Implement Beta Geometric conjugate prior for word segmentation
- Create Nepali segmentation dataset
- Create simple flask based web app
- Apply CI/CD pipeline
- Deploy on AWS Elastic Beanstalk
Please click here