top of page

Machine Learning

Utilizing a Bidirectional Long Short-Term Memory (LSTM) Model

Project Flow

Technologies: Google Colab, TensorFlow Keras, Python3

 

Retrieves data from database or csv hosted on github

​

Training Features:

  • 11 epochs, Restore Best Weights, Patience = 5

  • Adam Optimizer Training Rate = 0.01

  • Microbatch size = 1024

  • 2 data channels, 6 periodic waveforms

pseudoflow.png
Project Flow

LSTM vs Bidirectional LSTM

LSTMnets.png
LSTM vs BI-LSTM

Analysis Flow: Generating the 96-Point Forecast

Input

Raw Output

Apply Weights

Aggregate

Window and Score

Example Shown with Bitcoin (BTC) 

Input

BTC Input.png
Analysis Flow

Raw Output

BTC Raw Output.png
BTC Apply Weights.png

Apply Weights

Aggregate

BTC Aggregate.png
BTC PredvsActual.png

Window and Score

bottom of page