007-lab-001. learning rate,linear regression with not normalized data,linear regression with normalized data by MinMaxScaler()
<![CDATA[
import tensorflow as tf
tf.set_random_seed(777)
x_train_data=[[1,2,1],
[1,3,2],
[1,3,4],
[1,5,5],
[1,7,5],
[1,2,5],
[1,6,6],
[1,7,7]]
y_train_data=[[0,0,1],
[0,0,1],
[0,0,1],
[0,1,0],
[0,1,0],
[0,1,0],
[1,0,0],
[1,0,0]]
# We will evaluate our model by following test dataset
x_test_data=[[2,1,1],
[3,1,2],
[3,3,4]]
y_test_data=[[0,0,1],
[0,0,1],
[0,0,1]]
# [n,3]
X_placeholder_node=tf.placeholder("float",[None,3])
# [n,3]
Y_placeholder_node=tf.placeholder("float",[None,3])
# [n,3][?,?]=[n,3]
# [?,?]=[3,3]
W_variable_node=tf.Variable(tf.random_normal([3,3]))
# [3] is from [None,3]
b_variable_node=tf.Variable(tf.random_normal([3]))
h_f_softmax_node=tf.nn.softmax(tf.matmul(X_placeholder_node,W_variable_node)+b_variable_node)
cost_f_cross_entropy_node=tf.reduce_mean(-tf.reduce_sum(Y_placeholder_node*tf.log(h_f_softmax_node),axis=1))
# You try this model with small learning rate
gradient_descent_optimizer=tf.train.GradientDescentOptimizer(learning_rate=1e-10)\
.minimize(cost_f_cross_entropy_node)
prediction_after_argmax_node=tf.arg_max(h_f_softmax_node,1)
compare_prediction_and_label_node=tf.equal(prediction_after_argmax_node,tf.arg_max(Y_placeholder_node,1))
accuracy_node=tf.reduce_mean(tf.cast(compare_prediction_and_label_node,tf.float32))
with tf.Session() as sess_object:
sess_object.run(tf.global_variables_initializer())
for step in range(201):
if step%50==0:
# You use train dataset
loss_value,W_value,_=sess_object.run([cost_f_cross_entropy_node,W_variable_node,gradient_descent_optimizer]\
,feed_dict={X_placeholder_node:x_train_data,Y_placeholder_node:y_train_data})
print("step:",step,"loss_value:",loss_value,"W_value:",W_value)
# You let model to predict with test dataset
print("prediction:",sess_object.run(prediction_after_argmax_node,feed_dict={X_placeholder_node:x_test_data}))
# I calculate accuracy_node
print("accuracy_node: ",sess_object.run(accuracy_node,feed_dict={X_placeholder_node: x_test_data,Y_placeholder_node:y_test_data}))
# step: 0 loss_value: 5.7320304 W_value: [[ 0.8026957 0.6786129 -1.2172831]
# [-0.3051686 -0.3032113 1.508257 ]
# [ 0.7572236 -0.7008909 -2.108204 ]]
# step: 50 loss_value: 5.7320304 W_value: [[ 0.8026957 0.6786129 -1.2172831]
# [-0.3051686 -0.3032113 1.508257 ]
# [ 0.7572236 -0.7008909 -2.108204 ]]
# step: 100 loss_value: 5.7320304 W_value: [[ 0.8026957 0.6786129 -1.2172831]
# [-0.3051686 -0.3032113 1.508257 ]
# [ 0.7572236 -0.7008909 -2.108204 ]]
# step: 150 loss_value: 5.7320304 W_value: [[ 0.8026957 0.6786129 -1.2172831]
# [-0.3051686 -0.3032113 1.508257 ]
# [ 0.7572236 -0.7008909 -2.108204 ]]
# step: 200 loss_value: 5.7320304 W_value: [[ 0.8026957 0.6786129 -1.2172831]
# [-0.3051686 -0.3032113 1.508257 ]
# [ 0.7572236 -0.7008909 -2.108204 ]]
# prediction: [0 0 0]
# accuracy_node: 0.0
# @
import tensorflow as tf
import numpy as np
tf.set_random_seed(777)
xy_train_data=np.array([[828.659973,833.450012,908100,828.349976,831.659973],
[823.02002,828.070007,1828100,821.655029,828.070007],
[819.929993,824.400024,1438100,818.97998,824.159973],
[816,820.958984,1008100,815.48999,819.23999],
[819.359985,823,1188100,818.469971,818.97998],
[819,823,1198100,816,820.450012],
[811.700012,815.25,1098100,809.780029,813.669983],
[809.51001,816.659973,1398100,804.539978,809.559998]])
x_train_data=xy_train_data[:,0:-1]
y_train_data=xy_train_data[:,[-1]]
X_placeholder_node=tf.placeholder(tf.float32,shape=[None,4])
Y_placeholder_node=tf.placeholder(tf.float32,shape=[None,1])
W_variable_node=tf.Variable(tf.random_normal([4,1]),name='weight')
b_variable_node=tf.Variable(tf.random_normal([1]),name='bias')
h_f_softmax_node=tf.matmul(X_placeholder_node,W_variable_node)+b_variable_node
# This is simplified loss function
simplified_cost_f_node=tf.reduce_mean(tf.square(h_f_softmax_node-Y_placeholder_node))
gradient_descent_optimizer=tf.train.GradientDescentOptimizer(learning_rate=1e-5)
to_be_trained_node=gradient_descent_optimizer.minimize(simplified_cost_f_node)
sess_object=tf.Session()
sess_object.run(tf.global_variables_initializer())
for step in range(101):
if step%20==0:
cost_value_in_for,h_f_value,_=sess_object.run(
[simplified_cost_f_node,h_f_softmax_node,to_be_trained_node]\
,feed_dict={X_placeholder_node:x_train_data,Y_placeholder_node:y_train_data})
print("step:",step,"cost value:",simplified_cost_f_node,"\nPrediction:\n",h_f_value)
lab-07-3-linear_regression_min_max.py
import tensorflow as tf
import numpy as np
tf.set_random_seed(777)
# This method normalizes data
def MinMaxScaler(data):
numerator=data - np.min(data,0)
denominator=np.max(data,0) - np.min(data,0)
# noise term prevents the zero division
return numerator / (denominator + 1e-7)
xy_train_data=np.array([[828.659973,833.450012,908100,828.349976,831.659973],
[823.02002,828.070007,1828100,821.655029,828.070007],
[819.929993,824.400024,1438100,818.97998,824.159973],
[816,820.958984,1008100,815.48999,819.23999],
[819.359985,823,1188100,818.469971,818.97998],
[819,823,1198100,816,820.450012],
[811.700012,815.25,1098100,809.780029,813.669983],
[809.51001,816.659973,1398100,804.539978,809.559998]])
# very important. It does not work without it.
xy_train_data_scaled=MinMaxScaler(xy_train_data)
# array([[0.99999999, 0.99999999, 0. , 1. , 1. ],
# [0.70548491, 0.70439552, 1. , 0.71881782, 0.83755791],
# [0.54412549, 0.50274824, 0.57608696, 0.606468 , 0.6606331 ],
# [0.33890353, 0.31368023, 0.10869565, 0.45989134, 0.43800918],
# [0.51436 , 0.42582389, 0.30434783, 0.58504805, 0.42624401],
# [0.49556179, 0.42582389, 0.31521739, 0.48131134, 0.49276137],
# [0.11436064, 0. , 0.20652174, 0.22007776, 0.18597238],
# [0. , 0.07747099, 0.5326087 , 0. , 0. ]])
x_train_data=xy_train_data_scaled[:,0:-1]
y_train_data=xy_train_data_scaled[:,[-1]]
# [n,4]
X_placeholder_node=tf.placeholder(tf.float32,shape=[None,4])
# [n,1]
Y_placeholder_node=tf.placeholder(tf.float32,shape=[None,1])
# [n,4][?,?]=[n,1]
# [?,?]=[4,1]
W_variable_node=tf.Variable(tf.random_normal([4,1]),name='weight')
b_variable_node=tf.Variable(tf.random_normal([1]),name='bias')
hypothesis_function_node=tf.matmul(X_placeholder_node,W_variable_node)+b_variable_node
simplified_cost_f_node=tf.reduce_mean(tf.square(hypothesis_function_node-Y_placeholder_node))
gradient_descent_optimizer=tf.train.GradientDescentOptimizer(learning_rate=1e-5)
to_be_trained_node=gradient_descent_optimizer.minimize(simplified_cost_f_node)
sess_object=tf.Session()
sess_object.run(tf.global_variables_initializer())
for step in range(101):
if step%20==0:
cost_value,hypothesis_value,_=sess_object.run(
[simplified_cost_f_node,hypothesis_function_node,to_be_trained_node]\
,feed_dict={X_placeholder_node:x_train_data,Y_placeholder_node:y_train_data})
print(step,"Cost: ",cost_value,"\nPrediction:\n",hypothesis_value)
# 0 Cost: 0.10422044
# Prediction:
# [[0.91121554]
# [0.88515174]
# [0.8293996 ]
# [0.72472095]
# [0.82365894]
# [0.75689733]
# [0.6798932 ]
# [0.49118808]]
# 20 Cost: 0.104216896
# Prediction:
# [[0.91120636]
# [0.8851419 ]
# [0.8293911 ]
# [0.72471416]
# [0.82365113]
# [0.7568897 ]
# [0.6798872 ]
# [0.49118185]]
# 40 Cost: 0.10421333
# Prediction:
# [[0.9111972 ]
# [0.8851321 ]
# [0.82938266]
# [0.72470725]
# [0.8236433 ]
# [0.7568821 ]
# [0.6798811 ]
# [0.4911756 ]]
# 60 Cost: 0.10420978
# Prediction:
# [[0.91118807]
# [0.8851222 ]
# [0.8293742 ]
# [0.72470033]
# [0.8236356 ]
# [0.75687444]
# [0.6798751 ]
# [0.4911694 ]]
# 80 Cost: 0.104206234
# Prediction:
# [[0.91117895]
# [0.8851123 ]
# [0.82936573]
# [0.7246935 ]
# [0.82362777]
# [0.7568668 ]
# [0.67986906]
# [0.49116313]]
# 100 Cost: 0.10420268
# Prediction:
# [[0.91116977]
# [0.8851025 ]
# [0.82935727]
# [0.7246866 ]
# [0.82361996]
# [0.7568592 ]
# [0.679863 ]
# [0.4911569 ]]
]]>