numpy读写数据

numpy读写数据

1. 使用numpy读写ndarray

import numpy as np
aa=np.random.randint(1,4,size=(3,3))
print(aa)
np.save('ndarray_data.npy',aa)

bb=np.load('ndarray_data.npy')
print(bb)
print(type(bb))

显示结果为：

[[3 3 2]
 [1 3 1]
 [1 3 3]]
[[3 3 2]
 [1 3 1]
 [1 3 3]]
<class 'numpy.ndarray'>

2. 使用numpy保存和读取字典（value是torch.Tensor）

import numpy as np
import torch as t

a={}
w1=t.randn(3,3)
print(w1)
b1=t.randn(3,1)
print(b1)
w2=t.randn(3,3)
print(w2)
b2=t.randn(3,1)
print(b2)
a['conv1/weight']=w1
a['conv1/bias']=b1
a['conv2/weight']=w2
a['conv2/bias']=b2
np.save('aaa.npy',a)

print('save success')

tmp=np.load('aaa.npy')
print(type(tmp))
print(tmp[()].keys())

print('-'*20)
for key in tmp[()]:
    print(key,':')
    print(type(tmp[()][key]))
    print(tmp[()][key])
    print('-'*10)

显示结果为：

tensor([[ 0.5694,  0.8036, -0.6987],
        [-0.8518,  0.3729, -1.3868],
        [-0.0352, -0.7815,  0.3556]])
tensor([[1.0807],
        [2.4857],
        [0.5353]])
tensor([[-0.0558,  0.2604,  2.3809],
        [-0.1489, -0.3342,  1.7391],
        [-0.3890,  0.5136,  0.2696]])
tensor([[-2.5337],
        [-1.1947],
        [-0.4896]])
save success
<class 'numpy.ndarray'>
dict_keys(['conv1/weight', 'conv1/bias', 'conv2/weight', 'conv2/bias'])
--------------------
conv1/weight :
<class 'torch.Tensor'>
tensor([[ 0.5694,  0.8036, -0.6987],
        [-0.8518,  0.3729, -1.3868],
        [-0.0352, -0.7815,  0.3556]])
----------
conv1/bias :
<class 'torch.Tensor'>
tensor([[1.0807],
        [2.4857],
        [0.5353]])
----------
conv2/weight :
<class 'torch.Tensor'>
tensor([[-0.0558,  0.2604,  2.3809],
        [-0.1489, -0.3342,  1.7391],
        [-0.3890,  0.5136,  0.2696]])
----------
conv2/bias :
<class 'torch.Tensor'>
tensor([[-2.5337],
        [-1.1947],
        [-0.4896]])
----------

3. 使用numpy中的savez（）同时保存数组、字符串、字典数据
   参考链接：https://blog.csdn.net/lilong117194/article/details/83589510