Skip to content

Instantly share code, notes, and snippets.

@sreeragh-ar
Created August 17, 2018 08:07
Show Gist options
  • Save sreeragh-ar/77c084a6461d5fb821293efa199aed1d to your computer and use it in GitHub Desktop.
Save sreeragh-ar/77c084a6461d5fb821293efa199aed1d to your computer and use it in GitHub Desktop.
import tensorflow as tf
import numpy as np
import pandas as pd
from sklearn.datasets import load_boston
import matplotlib.pyplot as plt
boston=load_boston()
type(boston)
boston.feature_names
bd=pd.DataFrame(data=boston.data,columns=boston.feature_names)
bd['Price']=pd.DataFrame(data=boston.target)
np.random.shuffle(bd.values)
W0=tf.Variable(0.3)
W1=tf.Variable(0.2)
b=tf.Variable(0.1)
#print(bd.shape[1])
tf.summary.histogram('Weights', W0)
tf.summary.histogram('Weights', W1)
tf.summary.histogram('Biases', b)
dataset_input=bd.iloc[:, 0 : bd.shape[1]-1];
#dataset_input.head(2)
dataset_output=bd.iloc[:, bd.shape[1]-1]
dataset_output=dataset_output.values
dataset_output=dataset_output.reshape((bd.shape[0],1))
#converted (506,) to (506,1) because in pandas
#the shape was not changing and it was needed later in feed_dict
dataset_input=dataset_input.values #only dataset_input is in DataFrame form and converting it into np.ndarray
dataset_input = np.array(dataset_input, dtype=np.float32)
#making the datatype into float32 for making it compatible with placeholders
dataset_output = np.array(dataset_output, dtype=np.float32)
X=tf.placeholder(tf.float32, shape=(None,bd.shape[1]-1))
Y=tf.placeholder(tf.float32, shape=(None,1))
Y_=W0*X*X + W1*X + b #Hope this equation is rightly written
#Y_pred = tf.add(tf.multiply(tf.pow(X, pow_i), W), Y_pred)
print(X.shape)
print(Y.shape)
loss=tf.reduce_mean(tf.square(Y_-Y))
tf.summary.scalar('loss',loss)
optimizer=tf.train.GradientDescentOptimizer(0.0000000001)
train=optimizer.minimize(loss)
init=tf.global_variables_initializer()#tf.global_variables_initializer()#tf.initialize_all_variables()
sess=tf.Session()
sess.run(init)
wb_=[]
with tf.Session() as sess:
summary_merge = tf.summary.merge_all()
writer=tf.summary.FileWriter("Users/ajay/Documents",sess.graph)
epochs=10
sess.run(init)
for i in range(epochs):
s_mer=sess.run(summary_merge,feed_dict={X: dataset_input, Y: dataset_output}) #ERROR________ERROR
sess.run(train,feed_dict={X:dataset_input,Y:dataset_output})
#CHANGED
print("loss = ", sess.run(loss, feed_dict={X:dataset_input,Y:dataset_output}))
writer.add_summary(s_mer,i)
#tf.summary.histogram(name="loss",values=loss)
if(i%5==0):
print(i, sess.run([W0,W1,b]))
wb_.append(sess.run([W0,W1,b]))
print(writer.get_logdir())
print(writer.close())
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment