IoT applications are often based on time series data collected by the sensors in the IoT system. This means that the best deep learning models will often be designed with sequential data in mind, such as RNNs and LSTMs. In this assignment you will build an LSTM predictor using the household electricity consumption data from the previous two assignments to once again predict the Global Active Power usage based on simulated streaming data. You will be guided to build a simple model and then you will propose and implement changes to the model and analyze the effect that your changes have on the model's performance.
-
Create a repository under your GitHub account from this template: https://github.com/amarbut/aai-iot-lstm. Instructions can be found here. Make your repository private and add your instructor’s Github account as a collaborator.
-
Following the instructions in the jupyter notebook from the above Github template, build an LSTM model to predict Global Active Power.
-
In the sections provided in the jupyter notebook, make changes to the original LSTM model that you think will improve its performance.
-
Answer all questions as prompted in the jupyter notebook to analyze the performance of your models, the effects that your changes have on the predictions, and compare these models to the linear regression that you built in the previous assignment.
Assignment Materials:
Deliverables:
- When you have finished your code, print your notebook as a PDF and upload it to Blackboard.
- Commit your code and push the changes to GitHub so your instructor has access to the ipynb notebook files and any other code you create.