Side_projects

In my final project for Psychology 555: Detection and Discrimination, I simulated data with condition effect sizes set to 0 and compared false positive rates of different analysis techniques including a 1. point estimate method, 2. bayesian generalized linear model (GLM), and 3. bayesian hierarchical generalized linear model (GLMM). I found that while the point estimate and GLMM methods maintained an acceptable alpha level across multiple simulated experiments, the GLM false positive rate alpha levels was as high as 0.59 across conditions.

This program is designed to classify an input movie review as either negative or positive. This is done via a logistic regression model, decision tree model, and multilayer perceptron neural network model. There are several user options to run the code. Currently, the code is designed to output results for each model with raw or cleaned input text, with baseline or improved feature representations, and with different vectorizer decisions. While the features and textual representations are currently designed to be baseline vs. improved and raw vs. cleaned, the user can further designate which text representations or features to turn on and off. More information on this functionality is given in the comments of the main.ipynb script.