Switch branches/tags
Nothing to show
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
..
Failed to load latest commit information.
__pycache__
data
figures
project_proposal
project_report
ARIMA_routines.py
Benchmark_ARIMA_Results.ipynb
README.md
Time_Series_with_FNN.ipynb
Time_Series_with_LSTM.ipynb
final_project_write_up.pdf
network_architecture_FNN.png
project_proposal.pdf
requirements.txt
sunspot_FFNN.png

README.md

Udacity Capstone Project

This directory contain all code that was used for the Udacity Machine Learning Engineer Nanodegree Program. The folder figures contains all of the pdf figures generated by the codes. The links to the project proposal and the write-up of the final project are below.

The enviroments needed to run the program can be installed from the requirements.txt in anaconda using the following command:
$ conda create --name --file requirements.txt

The equations are generated from http://latex.codecogs.com

Project Summary

In this project we have taken data for time series data equation and we have trained two different neural network models, the feed forward neural network (FFNN) and the long-short term memory (LSTM) network, to generate a time series forecast given the value of the time series at the previous time step. In other words, we have generated a function equation such that

equation

where equation is similar to the equation value of the underlying time series data. The neural network models out performed the ARIMA and SARIMA benchmarks.

An example of the forecasts generated by feed forward neural network.

Included Code

  1. Benchmark ARIMA Results This jupyter notebook contains all of the code used to fit the ARIMA and SARIMA results used for benchmarking

  2. Time Series with FNN This jupyter notebook contains all of the code used to generate the time series results by using the Feed Forward neural network.

  3. Time Series with LSTM This jupyter notebook contains all of the code needed to generate the time series results by using LSTM network architectures.

  4. ARIMA routines This python code contains all of the fitting routines that are used by the code Benchmark_ARIMA_Results.ipynb

Datasets

The datasets that are included in the directory data directory are the following

  1. EURUSD_15m_BID_01.01.2010-31.12.2016
  • This is the Euro to USD exchange rate data from 2010-2016
  1. international-airline-passengers.csv
  • The data set of total number of international airline passengers from 1949-1960, in thousands.
  1. wolfer-sunspot-numbers-1770-to-1.csv
  • The number of sun spots observed from 1770-1869.
  1. total-annual-rainfall-in-inches-.csv
  • The total annual rainfall in London as a function of time from 1813-1912 in inches.