Exp-Time-Series-Tools/Practical-Deep-Learning-Applied-to-Time-Series

Practical Deep Learning resources for Time series analysis and forecasting

1. Practical DL applied to TS notebook collection:

This is the section where we can share the group's notebooks.

By data type:

• Univariate:
  • OliveOil
• Multivariate datasets:

By application:

• TS → images:
  • Unaltered time series:
  • Gramian Angular Field (GAF):
    • OliveOil
  • Wavelet transform:
• TS → text:
• TS → tabular data:

By NN model types:

• CNN models:
  • OliveOil
• RNN models:
  • Without ULMFIT pretraining:
  • With ULMFIT pretraining:
• Hybrid models:

By Approach:

• Transfer learning:
  • OliveOil
• Training from scratch:

By Target Type:

• Classification:
  • OliveOil
• Regression:

By # targets:

• Single label:
  • OliveOil
• Multi-label:

By # steps into the future:

• Single step:
  • OliveOil
• Multiple steps:

2. DL applied to TS resources: third party resources

This is the section where we can share useful thrid party resources.

Datasets:

• UEA & UCR Time Series Classification: contains 128 univariate and 30 multivariate datasets.
• UCI Machine Learning Repository: contains 452 ML datasets, with 86 time series (most multivariate datasets, both classifications and regression).

Libraries/ packages:

• Algorithms:

  • pyts: is a Python package for time series transformation and classification. It aims to provide state-of-the-art as well as recently published algorithms for time series classification.
  • tslearn: is a Python package that provides machine learning tools for the analysis of time series. This package builds on (and hence depends on) scikit-learn, numpy and scipy libraries.
  • PyDLM: A python library for Bayesian time series modeling.

• Databases:

  • arctic: high performance datastore for time series and tick data.
  • PyStore: fast data store for Pandas timeseries data.
  • Gnocchi: Gnocchi is an open-source time series database.

Third party notebooks/ repos:

Papers:

By NN model type:

• CNN models:

  • Tiled Convolutional Neural Networks:
    • Encoding Time Series as Images for Visual Inspection and Classification Using Tiled Convolutional Neural Networks (2015) - repo

• RNN models:

  • Dilated Recurrent Neural Networks:
    • Dilated Recurrent Neural Networks (2017) - repo

• Hybrid models:

  • Long Short Term Memory Fully Convolutional Network:

    • LSTM Fully Convolutional Networks for Time Series Classification (2017) - repo - * state of the art in 65/85 multivariate UCR datasets. Keras model.

    • Multivariate LSTM-FCNs for Time Series Classification (2018) - repo - * state of the art in many univariate datasets. Keras model.

Websites/ Blogs:

• Medium:
  • Pandas for time series data — tricks and tips