M.phill level deep learning assignment.(Python Expert is needed).

COM 762: Deep Learning and Its Application

Assignment 2 for MSc

Date set: 1st March 2021

Submission date due: 4th May 2021

Feedback date due: 24th May 2021

This assignment carries 50% of the coursework marks for the module. The complete

submission should be uploaded to the Assignment-2 folder in Blackboard by noon on

the due date.

The assignment provides an opportunity for students to develop a Long Short Term

Memory (LSTM) network for time series analysis using Python. LSTM networks is a

type of Recurrent neural networks. The big advantage of LSTM models is to

persistently accumulate propagation state over the input sequence, that is the output of

the model from one timestep is provided as an input in the subsequent timestep of the

model. This feature allows the model to make a prediction based on both the input for

the current timestep and prior knowledge derived in the previous timestep, resulting in

more stable prediction results.

The objective of this assignment is to develop and evaluate an LSTM model for multi-

step time series forecasting on selected data sources. The specific tasks include:

(a) define a many-to-many type sequence prediction problem, such as power

consumption forecast or solar energy forecast for next 3 days, etc.

(b) Design and develop an LSTM neural network with multiple input/output time steps

(c) This neural network should be composed of at least three hidden layers with two

different activation functions

(d) Conduct performance analysis of the LSTM model via training and testing on

appropriate data sources

The submission package includes

1. A report should consist of 5-7 pages in the IEEE format, covering the description

of tasks (a)-(d); and screenshots for tasks (d). Note that screenshots must be

captured in the PyCharm environment and submitted report must be a MS WORD

file.

2. The source code and the explanation of how to run the program.