Implementing and Evaluating Probabilistic Graphical Algorithms (Python)

  • The goal is to Identify and statistically measure individual writing habits as they develop.
  • Learned Bayesian Networks and Markov Networks from the discrete Cursive and Hand printed feature data of the School Children handwritings on the feature set of a particular word “and”.
  • Used Most Likely Independent Value Imputation method for finding missing and corrupted values. Pearson Chi Square test was used to find the correlation and log-loss method for direction between the features.
  • Inferenced random probabilistic queries from the constructed networks using Variable Elimination algorithm.
  • Developed Local Conditional Probability function by training a multiclass classifier (Used Logistic and Neural Net) for effective memory usage instead of creating Conditional tables.
  • Applied Sampling algorithms such as Ancestral and Gibbs Sampling and compared the samples with the given data using the mean and Entropy metrics.