我們使用最基礎的 Gradient Descent 方法來找到$y=f(x)=Wx+b$ 當中的 $W$ 和 $b$ , 用來預測 PokenMon 升級後的 CP 值

In [120]:

    

import numpy as np
import pandas as pd
pd = pd.read_csv("./pokemon.csv")
newcp = np.array(pd['cp_new'])
cp = np.array(pd['cp'])
cp = cp[:50]
newcp = newcp[:50]

import matplotlib
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline  



plt.scatter(cp, newcp)
plt.show()


    

In [121]:

    

import tensorflow as tf
# Initialize placeholders 
x = tf.placeholder(tf.float32)
y = tf.placeholder(tf.float32)

W = tf.Variable([1], dtype=tf.float32)
b = tf.Variable([1], dtype=tf.float32)
linear_model = W*x + b


    

In [122]:

    

squared_deltas = tf.square(linear_model - y)
loss = (tf.reduce_sum(squared_deltas))


    

In [123]:

    

optimizer = tf.train.GradientDescentOptimizer(0.00000001)
train = optimizer.minimize(loss)


    

In [124]:

    

sess = tf.Session()
init = tf.global_variables_initializer()
sess.run(init)
plt.figure(figsize=(30, 50))



index =0
predict_cp = sess.run(linear_model,{x:cp})

#Draw Chart
ax  = plt.subplot(6,4,1)
ax.scatter( cp,predict_cp)
ax.scatter( cp,newcp)  
for i in range(100):
    
    # Do Update !!!!
    sess.run(train, {x:cp, y: newcp})
    
    
    if i%5==0 :
        predict_cp = sess.run(linear_model,{x:cp})
        index +=1
        error = sess.run(loss, {x: cp,y: newcp})/5
        
        #Draw Chart
        ax  = plt.subplot(6,4,index+1)
        ax.set_title(str(i)+":" + str(error))
        ax.scatter( cp,predict_cp)
        ax.scatter( cp,newcp)
plt.show()


    

In [ ]: