Explore LSTM On Time Series Data (Without Any Code)

PoweredByTF 2.0 Challenge!

Purpose

The sole purpose for the development of this software was to make it easy for learners to understand how tuning several hyperparameters can effect the result of an LSTM (Long Short Term Memory) network on various types of time series data.

Dependencies

Python 3.x
Tensorflow
Matplotlib

Running the App

Install REQUIREMENTS.txt, by running pip install -r REQUIREMENTS.txt
Open app.py file
Set the hyperparameter values ( Dropout, Lag, Test Ratio, Max Epoch )
Select a preloaded dataset
Click Start
To reset the console click Reset

OS Support

The application has been tested on Windows and Linux platforms. In case of any issue, feel free to raise an issue.

Examples

Increasing Sales dataset

Sinusoidal Curve Dataset

Model Architecture

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
unified_lstm (UnifiedLSTM)   (None, LAG-1, 30)             3840      
_________________________________________________________________
dropout (Dropout)            (None, LAG-1, 30)             0         
_________________________________________________________________
unified_lstm_1 (UnifiedLSTM) (None, 30)                    7320      
_________________________________________________________________
dense (Dense)                (None, 1)                     31        
=================================================================
Total params: 11,191
Trainable params: 11,191
Non-trainable params: 0
_________________________________________________________________
None

Preloaded Datasets

Currently, the application supports 5 different datasets. We are going to add more datasets and probably improve the model in the next iteration of development. Contributions are welcomed.