Question

我决定使用一些实际示例来处理线性回归。为此，我下载了一个csv文件，该文件显示了1850年至2015年每个月的一些城市的温度数据。https://data.world/data-society/global-climate-change-data/workspace/file?filename=GlobalLandTemperatures%2FGlobalLandTemperaturesByMajorCity.csv读取该文件，计算年平均温度是正常的。但是，计算线性回归参数w存在问题。它等于nan。结果，无法建立回归线。你能告诉我错误是什么吗？预先感谢！

 import matplotlib.pyplot as plt
 import csv
 import tensorflow as tf

 tf.compat.v1.disable_eager_execution()
 def search_stat(start_year,finish_year):
   csv_path="GlobalLandTemperatures_GlobalLandTemperaturesByMajorCity.csv"
   with open(csv_path, encoding='utf-8', newline='') as csv_file:
   csvreader = csv.reader(csv_file)
   array= []
   end_year= str(finish_year)
   temperature=0
   n=-1

   #row[0]- column with dates in format YYYY-MM-DD
   #row[1]- column with monthly average temperature
   #row[3]- column with city names
   for row in csvreader:
      if (row[3]== 'Abidjan') and (start_year<=(int((row[0])[:4]))) and not 
      (row[0].startswith(end_year)):
         if (row[1]!='') and (row[1]!=','):
            temperature=float(row[1])
            array.append(temperature)
            n+=1
         else:
            array.append(array[n])
              n+=1
      if (row[0].startswith(end_year)):
          return array

def avr_tem_array(start_year,finish_year):
   array=[] #list with values in every months
   array= search_stat(start_year,finish_year)
   i=0
   sum=0
   k=0
   result_array=[]
   months=(finish_year-start_year)*12
   while(i<months):
      sum+=array[i]
      if k==11:
          result_array.append(sum/12) #average temperature for a year 
          sum=0
          k=0
          i+=1
          continue
      i+=1
      k+=1
  print(array)
  return(result_array)

array= []
array= avr_tem_array(1965,1995)#average annual temperatures 1965-1995
print(array)               

years=np.arange(1965.0,1995.0)
learning_rate=0.01
learning_epochs=100

X=tf.compat.v1.placeholder(tf.float32)
Y=tf.compat.v1.placeholder(tf.float32)

def model(X,w):
    return tf.compat.v1.multiply(X,w)

w= tf.compat.v1.Variable(0.0, name='weights')# parameter of regression
y_model=model(X,w)
cost=tf.square(Y-y_model)

main_fun=tf.compat.v1.train.GradientDescentOptimizer(learning_rate)
.minimize(cost)  

sess=tf.compat.v1.Session()
init=tf.compat.v1.global_variables_initializer()
sess.run(init)

for epoch in range(learning_epochs):
   for (x,y) in zip(years,array):
       sess.run(main_fun, feed_dict={X: x, Y: y})

w_learned=sess.run(w) #calculate final value of linear regression parametr
               #it turns out to be nan
sess.close()
y_lerned=w_learned*years #all values of y_learned is nan. We can't build line of regression
print(years)
print(w_learned)
print(y_lerned)
plt.scatter(years,array)
plt.plot(years,y_lerned) #building a regression line
plt.show()

不计算线性回归参数

0 个答案: