nekcht
Data Scientist with passion in Computer Vision and Big Data. Committed to leveraging technology to solve real-world problems.
Athens, Greece
Pinned Repositories
m4-competition-approaches
Project on implementing and comparing accepted approaches of the M4 Competition. The project was carried out as part of the curriculum for the course DI503 - Time Series Analysis and Applications, NKUA, Spring 2023.
apache-spark-evaluation
Evaluates the execution time differences between RDD (Resilient Distributed Datasets) and DataFrame data structures in Apache Spark. Also takes into account the file format being used, such as CSV or Parquet.
dynamic-time-warping
Dynamic Time Warping (DTW) implementation with and without locality constraint (window).
image-processing-algorithms
Python notebook demonstrating the practical application of various methods, including Canny Edge Detection, Harris Corner Detector, Hough Transform for Lines detection and Laplacian of Gaussian (LoG).
m4-timeseries-mlp
TimeSeries forecasting with a simple MLP model. It preprocesses the m4-competition dataset, encodes time-related info, scales data, and generates input sequences. The MLP, using TensorFlow/Keras, has layers tailored for the forecast horizon.
minhash-lsh-evaluation
Assessing MinHash LSH for text similarity. Compares with kNN using BART embeddings as ground truth. Involves data preprocessing, shingle creation, LSH experiments. Findings inform LSH's efficiency in document similarity tasks, enhancing understanding of LSH techniques.
ml-classic-scratch
Ordinary Least Squares, Ridge Regression, Expectation Maximization, Full Bayesian Inference, Bayes Classifiers, kNN, and MLP core algorithms from scratch. Some auxiliary functions are also used.
pneumonia-LogReg-HOG
Deep learning excels at learning hierarchical representations and handling large-scale data, but classical ML methods, when combined with HOG, a potent local shape and structure feature descriptor, can still deliver impressive performance, especially for specific image recognition tasks.
nekcht's Repositories
nekcht/apache-spark-evaluation
Evaluates the execution time differences between RDD (Resilient Distributed Datasets) and DataFrame data structures in Apache Spark. Also takes into account the file format being used, such as CSV or Parquet.
nekcht/dynamic-time-warping
Dynamic Time Warping (DTW) implementation with and without locality constraint (window).
nekcht/image-processing-algorithms
Python notebook demonstrating the practical application of various methods, including Canny Edge Detection, Harris Corner Detector, Hough Transform for Lines detection and Laplacian of Gaussian (LoG).
nekcht/m4-timeseries-mlp
TimeSeries forecasting with a simple MLP model. It preprocesses the m4-competition dataset, encodes time-related info, scales data, and generates input sequences. The MLP, using TensorFlow/Keras, has layers tailored for the forecast horizon.
nekcht/minhash-lsh-evaluation
Assessing MinHash LSH for text similarity. Compares with kNN using BART embeddings as ground truth. Involves data preprocessing, shingle creation, LSH experiments. Findings inform LSH's efficiency in document similarity tasks, enhancing understanding of LSH techniques.
nekcht/ml-classic-scratch
Ordinary Least Squares, Ridge Regression, Expectation Maximization, Full Bayesian Inference, Bayes Classifiers, kNN, and MLP core algorithms from scratch. Some auxiliary functions are also used.
nekcht/pneumonia-LogReg-HOG
Deep learning excels at learning hierarchical representations and handling large-scale data, but classical ML methods, when combined with HOG, a potent local shape and structure feature descriptor, can still deliver impressive performance, especially for specific image recognition tasks.