最近在學習numpy，pandas，tensorflow和pytorch，突然想試試各種不同的方法對矩陣的運算效率如何，以下是程式碼：

import tensorflow as tf
import pandas as pd
import torch
import time
import numpy as np
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'#mac Os 下有一個警告可以通過這個語句使之不顯

print('==============TensorFlow==============')

mat_1 = tf.constant(np.arange(12
 
).reshape(3,4).astype(np.float32))
mat_2 = tf.constant(np.arange(12).reshape(4,3).astype(np.float32))

start = time.time()
product = tf.matmul(mat_1,mat_2)
sess = tf.Session()
res = sess.run(product)
print(res)
end = time.time()
print('run rime = ',end-start)
print()


print('==============PyTorch=============='
 
)

x = torch.from_numpy(np.arange(12).reshape(3,4).astype(np.float32))
y = torch.from_numpy(np.arange(12).reshape(4,3).astype(np.float32))

start = time.time()
z = torch.mm(x,y)
print(z)
end = time.time()
print('run rime = ',end-start)
print()


print('==============Pandas==============')

x = pd.DataFrame(np.arange(12
 
).reshape(3,4).astype(np.float32))
y = pd.DataFrame(np.arange(12).reshape(4,3).astype(np.float32))

start = time.time()
z = x.dot(y)
print(z)
end = time.time()
print('run rime = ',end-start)
print()



print('==============NumPy==============')

x = np.arange(12).reshape(3,4).astype(np.float32)
y = np.arange(12).reshape(4,3).astype(np.float32)

start = time.time()
z = x.dot(y)
print(z)
end = time.time()
print('run rime = ',end-start)
print()

以下是執行結果：

可以看到pytorch確實比tensorflow對矩陣的運算效率要高不少。