This is an exploration into using TensorFlow to classify a data set.
The data set can be found here. It includes 17,898 records with 1,639 real pulsar examples and 16,259 negative examples caused by interference/noise. There are eight continuous variables and a binary class variable.
- Determine data set
- Build TensorFlow from source
- Classify data set
- Explore alternative classification techniques
- Write paper about methods tested
The paper will be written to NIPS standards, utilizing nips_2017.sty. The paper is located here at:
https://github.com/macattackftw/ResearchProject_HTRU2/blob/master/paper/paper.pdf
Final accuracy: 97.8687.
Loss function approaches zero very quickly.
To utilize this repository you will need the following:
After installing those you will need to download the dataset and place HTRU_2.csv in the same folder
as main.py. You will then be able to run main.py. If you wish to see my results
you can run the following from the main directory:
tensorboard --logdir model
This will will allow you to see all of the visualizations associated with this repository. Otherwise, if you wish to see how your model ran it should have saved under "model_X", where "X" is the run you wish to view. Command:
tensorboard --logdir model_X.