Something to read

In the previous article, we discussed the Conformer method, which was originally developed for weather forecasting. This is quite an interesting method. When testing the trained model, we got a pretty good result. But did we do everything right? Is it possible to get a better result? Let's look at the learning process. We are clearly not using the model forecasting the next most probable timeseries values for its intended purpose. By feeding the model input data from a timeseries, we trained it by propagating the error gradient from models using the prediction results. We started with the Critic's results.

RevIN — is a flexible, trainable layer that can be applied to any arbitrarily chosen layers, effectively suppressing non-stationary information (mean and variance of an instance) in one layer and restoring it in another layer of nearly symmetric position, such as input and output layers.

more...

Forecasting timeseries is one of the most important elements in building an effective trading strategy. When performing a trading operation in one direction or another, we proceed from our own vision (forecast) of the upcoming price movement. Recent advances in deep learning models, especially architecture-based Transformer models, have demonstrated significant progress in this area, offering a great potential for solving the multifaceted problems associated with long-term timeseries forecasting.

more...

Forecasting long-term timeseries is of specifically great importance for trading. The Transformer architecture, which was introduced in 2017, has demonstrated impressive performance in the areas of Natural Language Processing (NLP) and Computer Vision (CV). The use of Self-Attention mechanisms allows the effective capturing of dependencies over long time intervals, extracting key information from the context. Naturally, quite quickly a large number of different algorithms based on this mechanism were proposed for solving problems related to timeseries.

more...

In the previous article "Matrix Factorization: The Basics", I talked a little about how you, my dear readers, can use matrices in your general calculations. However, at that time I wanted you to understand how the calculations were done, so I didn't pay much attention to creating the correct model of the matrices.

more...

Long-term forecasting of time series is a long-standing problem in solving various applied problems. Transformer-based models show promising results. However, high computational complexity and memory requirements make it difficult to use the Transformer for modeling long sequences. This has given rise to numerous studies devoted to reducing computational costs of the Transformer algorithm.
Despite the progress made by Transformer-based time series forecasting methods based, in some cases they fail to capture the common features of the time series distribution. The authors of the paper "FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting" have made an attempt to solve this problem. They compare the actual data of a time series with its predicted values obtained from the vanilla Transformer. Below is a screenshot from that paper.

more...

The article presents the results of experiments on eight real data sets, according to which ATFNet shows promising results and outperforms other state-of-the-art time series forecasting methods on many datasets.

more...

In the previous article, we got acquainted with the ATFNet algorithm, which is an ensemble of 2 time series forecasting models. One of them works in the time domain and constructs predictive values of the studied time series based on the analysis of signal amplitudes. The second model works with the frequency characteristics of the analyzed time series and records its global dependencies, their periodicity and spectrum. Adaptive merging of two independent forecasts, according to the author of the method, generates impressive results.

more...