Live coding is the creative act of interactive code evaluations and multimodal assessments during the execution of a running program. In the context of music-making, the composer-programmer employs auditory percepts by means of appreciation and planning of present and future code evaluations. Thus, a weave of musical imagery and music listening facilitates the interplay between novel code evaluations and rigidly structured computations. Machine listening has been shown to be a prominent approach to control live coding and is typically correlated with perceptual assessments. Such developments have enabled the advent of statistical analysis of sequential dependencies between acoustical features during performance. Here, we will present a demo of a live coding session based on an OSC protocol binding between SuperCol￾lider and Python libraries for statistical learning. The system presented can perform window-based feature transformation of the extracted acoustical features, which consequently modify numerical parameters of the composition. A typical window size may be hundreds of milliseconds long, so that is multiple times the window size of feature extraction. This is done by storing the numerical values of the extracted features in plain text files. The approach used here aims to facilitate the understanding of the user and promote a human-centric approach to AI. Possible applications may be used for educational purposes and live music performances.