A real world application of a Recurrent Neural Network on a binary classification of time series data

Overview

What is this

This is a real world application of a Recurrent Neural Network on a binary classification of time series data. This project includes data cleanup, model creation, fitting, and testing/reporting and was designed and analysed in less than 24 hours.

Challenge and input

Three input files were provided for this challenge:

  • aigua.csv
  • aire.csv
  • amoni.csv (amoni_pred.csv is the same thing with integers rather than booleans)

The objective is to train a Machine Learning classifier that can predict dangerous drift on amoni.

Analysis procedure

Gretl has benn used to analyze the data.

Ideally, fuzzing techniques would be applied that would remove the input noise on amoni from the correlation with aigua.csv and aire.csv. After many hours of analysis I decided that the input files aire.csv and aigua.csv did not provide enough valuable data.

After much analysis of the amoni.csv file, I identified a technique that was able to remove most of the noise.

The technique has been implemented into the run.py file. This file cleanups up the data on amoni_pred.csv. It groups data by time intervals and gets the mean. It removes values that are too small. It clips the domain of the values. It removes noise by selecting the minimum values in a window slice. And (optionally) it corrects the dangerous drift values.

Generating the model

Once the file amoni_pred_base.csv has been created after cleaning up the input, we can move on to generating the model. Models are created and trained by the pred.py file. This file creates a Neural Network architecture with Recurrent Neural Networks (RNN). To be more precise, this NN has been tested with SimpleRNN and Long Short Term Memory (LSTM) layers. LSTM were chosed because they were seen to converge faster and provide better results and flexibility.

The input has been split on train/test sets. In order to test the network on fully unknown intervals, the test window time is non overlapping with the train window.

In order to allow prediction of a value, a window time slice is fed on to the LSTM layers. This window only includes past values and does not provide a lookahead cheat opportunity. The model is trained with checkpoints tracking testing accuracy. Loss and accuracy graphs are automatically generated for the training and testing sets.

Testing the models

After the models have been generated, the file test.py predicts the drift and dangerous values on the input data, It also provides accuracy metrics and saves the resulting file output.csv. This file can then be analysed with Gretl.

Performance

Our models are capable of achieving:

  • ~ 75% Accuracy on dangerous drifts with minimal time delays
  • ~ 80% Accuracy on drifts with minimal time delays

Moreover, with the set of corrections of the dangerous drift input values explained in previous sections, our model can achieve:

  • ~ 87% Accuracy on dangerous drifts with minimal time delays

Future Work / Improvements

Many improvements are possible on this architecture. First of all, fine tuning of the hyper parameters (clean up data set values, NN depth, type of layers, etc) should all be considered. Furthermore, more data should be collected, because the current data set only provides information for ~ 8 drifts. On top of that, more advanced noise analysis techniques should be applied, like fuzzing, exponential smoothing etc.

Other possible techniques

Yes, Isolation Forests are probably a better idea. But LSTM layers are cool :)

Show me some pictures

In blue, expected dangerous drift predictions. In orange the prediction by the presented model.

Screenshot1

Furthermore, with the patched dangerous drift patch:

Screenshot2

Owner
Josep Maria Salvia Hornos
Studying Business Management & Computer Science :D
Josep Maria Salvia Hornos
The official project of SimSwap (ACM MM 2020)

SimSwap: An Efficient Framework For High Fidelity Face Swapping Proceedings of the 28th ACM International Conference on Multimedia The official reposi

Six_God 2.6k Jan 08, 2023
Implementation of the Transformer variant proposed in "Transformer Quality in Linear Time"

FLASH - Pytorch Implementation of the Transformer variant proposed in the paper Transformer Quality in Linear Time Install $ pip install FLASH-pytorch

Phil Wang 209 Dec 28, 2022
Implementation of "StrengthNet: Deep Learning-based Emotion Strength Assessment for Emotional Speech Synthesis"

StrengthNet Implementation of "StrengthNet: Deep Learning-based Emotion Strength Assessment for Emotional Speech Synthesis" https://arxiv.org/abs/2110

RuiLiu 65 Dec 20, 2022
Animatable Neural Radiance Fields for Modeling Dynamic Human Bodies

To make the comparison with Animatable NeRF easier on the Human3.6M dataset, we save the quantitative results at here, which also contains the results of other methods, including Neural Body, D-NeRF,

ZJU3DV 359 Jan 08, 2023
Streaming over lightweight data transformations

Description Data augmentation libarary for Deep Learning, which supports images, segmentation masks, labels and keypoints. Furthermore, SOLT is fast a

Research Unit of Medical Imaging, Physics and Technology 256 Jan 08, 2023
Unified Pre-training for Self-Supervised Learning and Supervised Learning for ASR

UniSpeech The family of UniSpeech: UniSpeech (ICML 2021): Unified Pre-training for Self-Supervised Learning and Supervised Learning for ASR UniSpeech-

Microsoft 282 Jan 09, 2023
FusionNet: A deep fully residual convolutional neural network for image segmentation in connectomics

FusionNet_Pytorch FusionNet: A deep fully residual convolutional neural network for image segmentation in connectomics Requirements Pytorch 0.1.11 Pyt

Choi Gunho 102 Dec 13, 2022
Reinforcement Learning via Supervised Learning

Reinforcement Learning via Supervised Learning Installation Run pip install -e . in an environment with Python = 3.7.0, 3.9. The code depends on MuJ

Scott Emmons 49 Nov 28, 2022
Code and data of the ACL 2021 paper: Few-Shot Text Ranking with Meta Adapted Synthetic Weak Supervision

MetaAdaptRank This repository provides the implementation of meta-learning to reweight synthetic weak supervision data described in the paper Few-Shot

THUNLP 5 Jun 16, 2022
How to train a CNN to 99% accuracy on MNIST in less than a second on a laptop

Training a NN to 99% accuracy on MNIST in 0.76 seconds A quick study on how fast you can reach 99% accuracy on MNIST with a single laptop. Our answer

Tuomas Oikarinen 42 Dec 10, 2022
PyTorch implementation of NIPS 2017 paper Dynamic Routing Between Capsules

Dynamic Routing Between Capsules - PyTorch implementation PyTorch implementation of NIPS 2017 paper Dynamic Routing Between Capsules from Sara Sabour,

Adam Bielski 475 Dec 24, 2022
Code for Piggyback: Adapting a Single Network to Multiple Tasks by Learning to Mask Weights

Piggyback: https://arxiv.org/abs/1801.06519 Pretrained masks and backbones are available here: https://uofi.box.com/s/c5kixsvtrghu9yj51yb1oe853ltdfz4q

Arun Mallya 165 Nov 22, 2022
kapre: Keras Audio Preprocessors

Kapre Keras Audio Preprocessors - compute STFT, ISTFT, Melspectrogram, and others on GPU real-time. Tested on Python 3.6 and 3.7 Why Kapre? vs. Pre-co

Keunwoo Choi 867 Dec 29, 2022
Neon-erc20-example - Example of creating SPL token and wrapping it with ERC20 interface in Neon EVM

Example of wrapping SPL token by ERC2-20 interface in Neon Requirements Install

7 Mar 28, 2022
Code for the published paper : Learning to recognize rare traffic sign

Improving traffic sign recognition by active search This repo contains code for the paper : "Learning to recognise rare traffic signs" How to use this

samsja 4 Jan 05, 2023
Anomaly detection in multi-agent trajectories: Code for training, evaluation and the OpenAI highway simulation.

Anomaly Detection in Multi-Agent Trajectories for Automated Driving This is the official project page including the paper, code, simulation, baseline

12 Dec 02, 2022
Code for Fully Context-Aware Image Inpainting with a Learned Semantic Pyramid

SPN: Fully Context-Aware Image Inpainting with a Learned Semantic Pyramid Code for Fully Context-Aware Image Inpainting with a Learned Semantic Pyrami

12 Jun 27, 2022
This repo generates the training data and the model for Morpheus-Deblend

Morpheus-Deblend This repo generates the training data and the model for Morpheus-Deblend. This is the active development repo for the project and as

Ryan Hausen 2 Apr 18, 2022
Technical experimentations to beat the stock market using deep learning :chart_with_upwards_trend:

DeepStock Technical experimentations to beat the stock market using deep learning. Experimentations Deep Learning Stock Prediction with Daily News Hea

Keon 449 Dec 29, 2022