https://kirankumar.ai/blog/ Recent content on KK's Notes Hugo -- gohugo.io en-us Fri, 13 May 2022 20:20:00 +0000 https://kirankumar.ai/blog/posts/reinforcement_learning/ Fri, 13 May 2022 20:20:00 +0000 https://kirankumar.ai/blog/posts/reinforcement_learning/ A reinforcement learning model works with an agent or actor, on a set of states and actions optimizing for a reward. A policy determines the action in a state generating a reward; the goal of reinforcement learning is to learn this policy. RL algorithms have enjoyed a great deal of success in games such as atari, chess or go. https://kirankumar.ai/blog/posts/data_science_ethics_governance/ Fri, 14 Jan 2022 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_ethics_governance/ Ethics and governance. In the last post, we discussed monitoring through controls and baselines, watching a model after it ships so that drift or quiet degradation gets caught before it costs real money. This final post in the series asks a different question: is the model doing the right thing, and who is answerable if it is not. It returns one last time to mysurance, the insurance fraud model that opened the series, as a capstone case study in fairness, bias, and governance. https://kirankumar.ai/blog/posts/data_science_monitoring/ Fri, 12 Nov 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_monitoring/ Monitoring through controls and baselines. In the last post we discussed establishing a baseline once a model moves from a single experiment to full scale deployment. A baseline is a snapshot of how a model is expected to behave the day it goes live. This post is about what happens after that day, once the model is out in the world and the world keeps moving. https://kirankumar.ai/blog/posts/data_science_baseline/ Fri, 24 Sep 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_baseline/ Establishing a baseline. In a previous post, generalizing or scaling an experiment from a single pilot to many stores, regions, or customers is discussed. Once a solution is expected to run broadly, the question changes from whether an effect exists to whether the model built to exploit it is actually worth its cost. Answering that question requires a baseline, a simple reference prediction that a sophisticated model must clearly beat before its added complexity is justified. https://kirankumar.ai/blog/posts/data_science_scaling_experiment/ Fri, 13 Aug 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_scaling_experiment/ Generalizing or scaling the experiment (1 to n). A model validated in one controlled setting is not the same thing as a model that works everywhere. This post covers external validity, Simpson's paradox, the infrastructure concerns that only appear at scale, and a staged rollout approach, illustrated with mysurance scaling its fraud model across product lines and regions. https://kirankumar.ai/blog/posts/data_science_qualitative_modeling/ Fri, 25 Jun 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_qualitative_modeling/ Qualitative modeling. In the last two posts we built numeric models: a fitted risk score for hospital readmission and a pass on reducing its error through numerical analysis. This post is a deliberate turn away from that path, covering rule based systems, expert heuristics, thematic coding of qualitative feedback, and when a human reviewed rubric is the safer choice over a trained classifier. https://kirankumar.ai/blog/posts/data_science_numerical_analysis/ Fri, 21 May 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_numerical_analysis/ Numerical analysis for error reduction. In the previous post, we picked a model class for a hospital readmission problem. This post covers the numerical work that comes after picking a model class: gradient based optimization, why the loss function matters, regularization, and calibration. https://kirankumar.ai/blog/posts/data_science_modeling_patterns/ Fri, 16 Apr 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_modeling_patterns/ Modeling of patterns for prediction. A pattern has been found and validated well enough to trust for now, and the task is to turn it into a model that can produce a prediction for a record it has never seen: choosing model complexity, generalization, and the bias and variance tradeoff. https://kirankumar.ai/blog/posts/data_science_hypothesis_exploration/ Fri, 05 Mar 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_hypothesis_exploration/ Hypothesis driven exploration. In the last post, we looked at collection and analysis of data for pattern discovery using a ride hailing example. This post is about resisting the urge to act on a pattern immediately, and about what it takes to trust a pattern found while exploring rather than one set out to test. https://kirankumar.ai/blog/posts/data_science_collection_analysis/ Fri, 22 Jan 2021 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_collection_analysis/ Collection and analysis of data for pattern discovery. This post turns to data that was not collected to answer one designed question at all, but gathered passively through logging or instrumentation, and then explored to discover patterns worth turning into hypotheses. https://kirankumar.ai/blog/posts/data_science_controlling_factors/ Fri, 11 Dec 2020 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_controlling_factors/ Controlling other factors to observe primary effect. This post covers how to identify confounding factors and control for them, so that an observed difference can be attributed to the change under test and not to something else riding along with it. https://kirankumar.ai/blog/posts/data_science_experimental_design/ Fri, 06 Nov 2020 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_experimental_design/ Experimental design (0 to 1). This post is about the first experiment, the smallest and cheapest test that can tell us whether an idea is worth pursuing further before any scaling, tooling, or automation gets built around it. https://kirankumar.ai/blog/posts/data_science_latent_factors/ Fri, 18 Sep 2020 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_latent_factors/ Latent or unobservable factors. Not everything that drives an outcome can be measured directly. This post covers latent variables, the things we can only infer through proxies, using subscription cancellation for a media streaming product as the worked example. https://kirankumar.ai/blog/posts/data_science_process/ Fri, 07 Aug 2020 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_process/ Process of Data Science There are various processes which determine success in every field. A surgeon, pilot and scientist undergo many hours of training, spanning years, to practice. The hours of training are generally used to establish skills towards automaticity. The development of automaticity enables quick decisions and efficient management of tough situations. This idea of automaticity is enshrined in many theories of cognitive science such as the two process model by Schneider and Shiffrin in 1977. https://kirankumar.ai/blog/posts/data_science_measurement/ Fri, 07 Aug 2020 14:00:00 +0000 https://kirankumar.ai/blog/posts/data_science_measurement/ Measurement variables In a previous post, the process of data science and forming an hypothesis is discussed. A hypothesis is the relevant to align a business objective to a data science problem. The hypothesis provides a “big-picture” view of the issues which need to considered in further steps of addressing a data science problem. https://kirankumar.ai/blog/posts/attention/ Sun, 13 Jan 2019 22:27:46 +0000 https://kirankumar.ai/blog/posts/attention/ The digital era and the digital economy are driven by industrial disruptions such as Amazon - e-commerce, Facebook — social interaction, Netflix — entertainment, Uber — transportation and many such ventures. The key aspects of a digital economy are scale and speed. Scale conveys these digital ventures capture a user base of millions (if not billions). Speed conveys their ability to grow and adapt to the market at a rapid scale. https://kirankumar.ai/blog/about/ Wed, 19 Dec 2018 18:47:08 +0000 https://kirankumar.ai/blog/about/ This log is to capture notes and explorations while traversing the scope of experimental research, analysis, statistical modeling and machine learning. I am a Phd candidate at Indiana University researching and developing statistical models to understand different aspects of human attention and decision making. I have a Master’s degree in computer science and hands on experience in software development for over 3 years which has enabled me to create robust and efficient algorithms to interpret and analyze behavioral, image and textual data.