https://github.com/onmax/bachelor-thesis-code
Bachelor's thesis about comparing different models using RNN for forecasting bike demand in Chicago
https://github.com/onmax/bachelor-thesis-code
bikesharing forecasting neural-networks recurrent-neural-networks
Last synced: about 1 year ago
JSON representation
Bachelor's thesis about comparing different models using RNN for forecasting bike demand in Chicago
- Host: GitHub
- URL: https://github.com/onmax/bachelor-thesis-code
- Owner: onmax
- Created: 2020-12-12T16:44:07.000Z (over 5 years ago)
- Default Branch: main
- Last Pushed: 2021-02-12T18:49:45.000Z (over 5 years ago)
- Last Synced: 2025-03-23T22:21:22.539Z (about 1 year ago)
- Topics: bikesharing, forecasting, neural-networks, recurrent-neural-networks
- Language: Jupyter Notebook
- Homepage:
- Size: 823 MB
- Stars: 0
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Forecasting bike demand
This projects aims to create differente neural networks that predict the bike usage in a given set of intervals. The code uses datasets with each row as interval and each column with different data about that interval. This data can be: `hour`, `day_of_month` or `month` with the real data about the stations that are about to be predicted of the previous intervals. It also uses a sliding windows which a generator similar to the one that you can see in [this tutorial](https://www.tensorflow.org/tutorials/structured_data/time_series#data_windowing).
## Paper
The final paper can be viewed in [this repository](https://github.com/onmax/bachelor-thesis-paper/tree/main/data) where it shows more info, the results and some theory about neural networks and RNN.
## Models created
In this project 5 models have been developed:
- Baseline: It just return the prediction with the same data as one week before
- Dense: It is just two layers of dense units
- SimpleRNN: A SimpleRNN layer with a dense one
- LSTM: A LSTM layer with a dense one
- Autoregressive: A model that uses its own prediction as input. Also it relays heavily on [this tutorial](https://www.tensorflow.org/tutorials/structured_data/time_series#advanced_autoregressive_model).
## WindowGenerator
It follows the same logic as in the [tutorial of Tensorflow](https://www.tensorflow.org/tutorials/structured_data/time_series#data_windowing).
It can create subdatasets given a dataset. Given a dataset of 7 intervals, and the window configuration as 3 inputs and 1 output will generate 4 windows. Graphically:
## Original dataset
The original dataset is from the company [Divvy in Chicago](https://divvy-tripdata.s3.amazonaws.com/index.html). Some modifications have been developed and part of the data is being saved in this repository as [byte files](./data/parts).
## Library used
This project depends mainly on:
- Tensorflow and Keras as Machine Learning library
- Pandas and Numpy as data structures libraries
- Matplotlib for plotting with SciencePlot theme
----------
#### This project has been my bachelor's thesis.
The paper written for this paper can be seen at: [onmax/bachelor-thesis-paper](https://github.com/onmax/bachelor-thesis-paper).