Our skills in data science and machine learning include knowledge and experience in using such languages as Python, C#, C++ , and R, such libraries as Keras, TensorFlow, Matplotlib, Pandas, NumPy, SymPy, SciPy, Scrapy, Selenium, Requests, BS4, LXML, JSON, GGplot, OpenCV2, LightGBM-20, Dplyr, Forecast, GGplot2, Readr, Tseries, Tidyr, Scikit-Learn. I have experience working with databases (SQLite, PostgreSQL, MySQL) and Git (GitHub, Bitbucket).
Our competence includes data collection, processing, analysis, creation and optimization of machine learning models, and data placement on a server or database.
Extracting, processing, and analyzing trading analytical data for cryptocurrencies, everyday collecting of statistics, and analysis of trends, dependencies, patterns, and anomalies. The database collected: 41200 analytical reports.
'BTC' , 'ETH' , 'BNB' , 'SOL' , 'ADA' , 'XRP' , 'DOT' , 'DOGE' , 'AVAX' , 'LTC' , 'MATIC' , 'ALGO' , 'BCH' , 'EGLD' , 'VET' , 'XLM' , 'ICP' , 'TRX' , 'ATOM' , 'THETA' , 'FIL' , 'ETC' , 'FTM' , 'HBAR' , 'NEAR' , 'HNT' , 'XTZ' , 'XMR' , 'FLOW' , 'EOS' , 'RVN' , 'KLAY' , 'ONE' , 'XEC' , 'SRM' , 'KSM' , 'ONT' , 'NEO' , 'ZEC' , 'STX' , 'WAVES' , 'AR' , 'DASH' , 'CELO' , 'IOTX' , 'BTG' , 'XEM' , 'DCR' , 'QTUM' , 'ROSE' , 'MINA' , 'ZIL' , 'SC'
Binance, TradingView, min-api.cryptocompare, blockchaincenter.net, alternative.me, marketwatch.com, kitco.com, and other APIs and websites.
indexes, metrics, indicators, strategies signals, and others. Analytical parameters after processing and calculations: 101.
Collected data were analyzed with R libraries. Technologies used: Pandas, Matplotlib, SeaBorn, Requests, HTTPAdapter, LXML, Binance API, TradingView API, NumPy, SymPy, Dplyr, Forecast, Tidyr.
Powerful artificial neural networks for predicting the course of cryptocurrencies. Technologies used: NumPy, Matplotlib, Keras, TensorFlow. It is developed 5 configurations for neural networks. One configuration as an example: neurons activation function - ELU, layers number - 53, input parameters- 101, output parameters - 4, layers type – Dense, optimizer - Adam, algorithm convergence step - 0.0000195, loss function – MAE, metrics – MAPE, Iterations of training: 600 epochs, batch size: 30. Percent of correct predictions: 72%.
Large ARM model for exchange course predicting. Based on statistical data that include indicators, indexes, support and resistance levels, and history of course changing. Technologies used: TensorFlow, NumPy, SymPy.
The dashboard is created to display the stock glass as a depth chart, recognize and display accumulations of buy and sell orders (bids/asks), and determine local and global support and resistance levels.
The scraper searches for a certain topic in the TikTok social network, review the most popular posts by topic, saves their content with the statistics of the post and its author: the number of views, comments, and likes, uploads videos, saves the title of the video, hashtags, and comments. The collected information was analyzed to determine trends in the social network.
ETL system for Australian commercial real estate. The system includes 3 Python scrapers for extracting all information about real estate objects and agencies on Australian real estate websites. The parsers save text, numeric data, images, and video and load them into the PostgreSQL database. The collected data is intended for the analysis of the real estate market.
Real estate prices estimation with a neural network. The system was created to estimate prices for real commercial estate objects by multiple parameters.
Car price estimation for car trader on auto.ria. Developed software needs to estimate car prices by such parameters as the car’s developer, model, year of manufacture, mileage, tank volume, and fuel type.
Computer vision script for recognizing handwritten numbers.
Python ANN framework created for automation of compiling neural networks. Framework based on NumPy library.
The research was conducted on the automation of machine learning. The program generated 2000 neural networks of different configurations with different training settings with different amounts of training data. Valuable statistical data were collected on the performance of these neural networks when tested on the validation sample. The resulting dataset was thoroughly analyzed. The obtained conclusions were used in the development of all neural networks appearing in this portfolio.
The neural network is written in python using Keras, NumPy, and Matplotlib libraries for approximating polynomial regression dependencies for predicting.
C# engine for GNN (Graph neural networks). Functional for compiling GNN, preparing datasets for training, and functional for GNN optimization.
Python optimizer based on genetic algorithm. Created to recognize dependencies and approximate them for predicting.
The indicator system is written in the PineScript language to display support and resistance channels.
The indicator system written in PineScript is developed for dynamic display of support and resistance blocks.
The indicator system written in the PineScript language is created to display local and global volume clusters.
A Python script that draws random trees using a mathematical algorithm. This technology is called L-systems and is used mostly in graphics to depict plants.