For this assignment, you’ll be using a small (for real-world data) data set to analyze astrophysics data.
Pulsars are a rare type of neutron star that produce pulses (thus the name) of electromagnetic radiation in a regular cycle. However, there are many other things that also produce pulses of radiation, and many astronomical objects that appear to be candidate pulsars are in fact something else (false positives). Most of these are because of radio frequency interference (RFI) or noise.
Because the pulses vary slightly from one cycle to another, a candidate is observed for an extended period and the data from it is statistically profiled. Your data file contains information on 17,898 candidates, of which 1,639 are pulsars. Each row lists 8 variables, statistical summaries of the data from the star, and 1 outcome variable, the class (0 = RFI, 1 = pulsar). Your task is to use TensorFlow to construct a neural network to analyze this data.
All input data is numeric, floating point, and does not need to be rescaled or recoded. (You’re welcome.) Use a 70/15/15 split for the training, test, and validation sets. You should be able to achieve a fairly high level of classification with a fairly simple network; you will probably need no more than 2 hidden layers. But experiment and see what works.
Write up a short report discussing:
- The configuration of your network, along with a brief overview of your code
- Your cross-validation strategy
- Summarize your results
- Any further questions you’d like to discuss, ideas for extending this further, etc.
Submit your report and your source code by the deadline.
- Mean of the integrated profile.
- Standard deviation of the integrated profile.
- Excess kurtosis of the integrated profile.
- Skewness of the integrated profile.
- Mean of the DM-SNR curve.
- Standard deviation of the DM-SNR curve.
- Excess kurtosis of the DM-SNR curve.
- Skewness of the DM-SNR curve.
This data came from https://www.kaggle.com/pavanraj159/predicting-a-pu…