Improving the Predictions of Linear Models

In the prior recipes, we have noted that the number of parameters we are fitting far exceeds the equivalent linear models. In this recipe, we will attempt to improve our logistic model of low birthweight with using a neural network.

Getting ready

For this recipe, we will load the low birth-weight data and use a neural network with two hidden fully connected layers with sigmoid activations to fit the probability of a low birth-weight.

How to do it

1. We start by loading the libraries and initializing our computational graph:

   import matplotlib.pyplot as plt
   import numpy as np
   import tensorflow as tf
   import requests
   sess = tf.Session()

2. Now we will load, extract, and normalize our data just like as in the prior recipe, except that we are going to using the low birthweight indicator variable as our target instead of the actual birthweight:

   birthdata_url = 'https://www.umass.edu/statdata/statdata/data/lowbwt.dat''
   birth_file = requests.get(birthdata_url)
   birth_data...