Thresher - THRESHold EvaluatoR for Python, Spark and Ray
https://github.com/oskar-j/thresherOpen source · Python library & CLI · pip install thresher-py
A classifier gives you probabilities, and turning them into yes/no answers needs a cut-off - which almost
every pipeline leaves at the default 0.5. Thresher takes your predict_proba-like scores and the
ground truth, and returns the threshold that classifies the highest fraction of samples correctly.
Seven search algorithms (from an exact sweep to faster histogram and search-based ones), support for
pandas, Apache Spark and Ray, usable as a library or straight from the terminal.
Documentation: thresher.readthedocs.io.
FastCat - fast lookups of Wikipedia categories
https://github.com/oskar-j/fastcatOpen source · Python & Redis · pip install fastcat
A Python library for quickly looking up broader/narrower relations between Wikipedia categories locally, from DBpedia SKOS dumps loaded into Redis - without hammering the Wikipedia API. Started as a fork of Ed Summers' original tool; I ported it to Python 3, added support for 7+ Wikipedia languages, made loading ~20x faster with pipelined Redis writes and added a new Databus engine. Co-maintained on PyPI.
Walsh-Hadamard transform image codec
https://github.com/oskar-j/walsh-hadamard-transformOpen source · Python & NumPy
A complete image codec built on the Walsh-Hadamard transform - a transform from the 1920s that needs nothing but additions and subtractions. Pure Python and NumPy, every step exact (the same picture compresses to the same bytes on every machine), seven image file formats in and out, DCT and Haar transforms on board to compare against, and more than seven hundred tests. Works as a CLI and as a library.
Curated awesome lists for AI and ML
https://github.com/oskar-j?tab=repositories&q=awesomeOpen source · curated lists, 150+ GitHub stars in total
Lists of tools, libraries, datasets and research that I collect and maintain: awesome-ai-spotter (detecting AI-generated text, images, video and audio), awesome-image-coloring (AI-powered image colorization, featured in awesome-awesomeness), awesome-auto-ml (AutoML frameworks), awesome-sim-software (multi-agent simulation frameworks) and awesome-text-ml (ML for text data).
plwordnet-reader - Polish WordNet for Python
https://github.com/oskar-j/plwordnet-readerOpen source · Python, NLP
A Python package for iterating over the Polish wordnet ("Słowosieć", v3.2).
ADA - Adaptive anomaly detection
https://www.deep.bi/#retailCommercial, closed source
Fraud detection in retail stores/restaurants. Deep.BI uses AI pattern detection to analyze POS transactional data and detect employee fraud. Solution doesn't just look at the transactions one-by-one; algorithm analyzes employee history and location aggregated data, along with in-transaction data about item deletes, item additions, refunds and employee profiles, through neural networks and machine learning models to spot irregularities invisible to the human eye.
Skip Tracing & Cyber Tracking
https://web.archive.org/web/2018/https://togetherdata.com/Commercial, closed source
A dunning system, which boosts data collection capacity. Our solutions enabled to increase the efficiency of our clients data collection strategy. We automated the manual process of skip tracing and offer high performance along with capacity service that collects information about given individuals from multiple data sources, including mediasphere.
BigSim for maritime
https://www.scensei.com/Commercial, closed source
I worked on multi-agent big data simulation and optimization problem for the maritime industry. Our system optimized workflow of major ports in the America. Specialized operations research made for better performance of the port, higher ratio of found contraband, and finally, lowering costs. Produced recommendations were customer-tailored for every maritime port.
Content targeted advertising
https://web.archive.org/web/2021/https://www.applica.ai/Commercial, closed source
Context-aware targeting of online advertisement through using advanced NLP and AI techniques.
HYIP - simulation of High Yield Investment Programs
https://github.com/ResearchGeek/hyip-simulationAcademic, open source
A multi-agent simulation of High Yield Investment Program (Ponzi-like) strategies, with a scraper of HYIP monitoring sites and an R analysis of the results. Basis of the paper "Looking for big money in the grey zone" (Computer Science, 2015) - see publications.
Forecasting views for Youtube content
https://github.com/ShowMeModel/forecaster-viewsOpen source
Example of regression with caret framework.
Analysis of PM2.5 air contamination for the US data
https://github.com/ShowMeModel/pm25-us-analysis/Open source
Sample analysis of NEI public data about air contamination in USA.
Analyzing open source teams of developers located on the GitHub portal
https://github.com/wikiteams/Academic, open source
"WikiTeams" is a research grant from the National Science Centre, Poland (NCN). The research was conducted at the Polish-Japanese Academy of Information Technology. The main goals were: a) Modeling and evaluating quality of virtual teams of programmers. b) Creating recommendations for collaborative innovation networks. c) Investigating emergent task allocation strategies. Results were published in several papers - see publications.