# indexing
a = np.arange(1, 13).reshape(3, 4)
b = a[:2, 1:3]
print("a:\n", a, '\n')
print("b:\n", b, '\n')
b[0, 0] = 100
print("a_aft_changeb:\n", a, '\n') # b取a的部分值去操作，b改變相應的a處值也會變，注意=和copy的區別
# 用整數方式有可能得到降維的陣列，而用切片方式得到的陣列永遠是原陣列的子陣列
c = np.arange(1, 13).reshape(3, 4)
c_1 = c[1, :]
c_2 = c[1:2, :]
print("c_1:\n", c_1, '\n') # (4,)
print("c_2:\n", c_2, '\n') # (1,4)
# 陣列索引陣列
twod = np.arange(1, 10).reshape(3, 3)
twod_index = twod[[0, 2], [2, 1]] # (0,2)(2,1)
print("twod_index:\n", twod_index, '\n')
threed = np.arange(1, 9).reshape(2, 2, 2)
threed_index = threed[[0, 1, 0], [0, 0, 0], [1, 1, 0]] # (0,0,1)(1,0,1)(0,0,0,)
print("threed_index\n", threed_index, '\n')
# 用整數陣列給原陣列的元素加上乙個數值
d = np.arange(1, 13).reshape(4, 3)
d[np.arange(3), np.arange(3)] += 10 # (0,0)(1,1)(2,2)上的值均加10
print("d:\n", d, '\n')
# 選出陣列中符合某種特定條件的元素
e = np.arange(1, 7).reshape(3, 2)
bool_idx = (e > 2)
print("bool_idx:\n", bool_idx, '\n')
print(e[bool_idx], '\n')
f = np.array([[1, 3], [3, 4]])
print(f[f > 2], '\n')
x = np.arange(10)
print("x[1:7:2]:\n", x[1:7:2], '\n')
print("x[-2:10]:\n", x[-2:10], '\n')
print("x[-3:3:-1]:\n", x[-3:3:-1], '\n')
y = np.arange(1, 7).reshape(2, 3, 1)
print("y[1:2]:\n", y[1:2], '\n')
print("y[1]:\n", y[1], '\n')
print("y[..., 0]:\n", y[..., 0], '\n')
print("y[:,np.newaxis,:,:]:\n", y[:,np.newaxis,:,:], '\n') # 213
1rows = np.array([0, 3], dtype=np.intp)
cols = np.array([0, 2], dtype=np.intp)
z = np.arange(12).reshape(4, 3)
rows_new = rows[:, np.newaxis] # [[0], [3]]
print("z[rows_new, cols]:\n", z[rows_new, cols], '\n')
print("z[np.ix_(rows, cols)]:\n", z[np.ix_(rows, cols)], '\n') # 這兩結果一樣
print("z[1:2, 1:3]:\n", z[1:2, 1:3], '\n')
print("z[1:2, [1, 2]]:\n", z[1:2, [1, 2]], '\n') # 這兩結果一樣
xx = np.array([[1., 2.], [np.nan, 3.], [np.nan, np.nan]])
print("xx[~np.isnan(xx)]:\n", xx[~np.isnan(xx)], '\n') # all not nan
yy = np.array([[1., -1.], [-2., 3]])
yy[yy < 0] += 100 # add a constant to all negative elements
print("yy:\n", yy, '\n')
zz = np.array([[0, 1], [2, 2], [1, 1]])
rowsum = zz.sum(-1) # shape=(3,)
print("zz[rowsum <= 2, :]:\n", zz[rowsum <= 2, :], '\n')
zz = np.array([[0, 1], [1, 1], [2, 2]])
rowsum_dims = zz.sum(-1, keepdims=true) # shape=(3,1)
# print(rowsum_dims.shape)
# print("zz[rowsum_dims <= 2]:\n",zz[rowsum_dims <= 2], '\n') # python3報錯，python2 warning
xy = np.arange(12).reshape(4, 3)
rows = (xy.sum(-1) % 2) == 0 # 和是否整除2
print("rows:\n", rows, '\n')
cols = [0, 2]
print("xy[np.ix_(rows, cols)]:\n", xy[np.ix_(rows, cols)], '\n')
rows1 = rows.nonzero()[0]
print("rows1:\n", rows1, '\n')
print("xy[rows1[:, np.newaxis], cols]:\n", xy[rows1[:, np.newaxis], cols], '\n')
a:
[[ 1 2 3 4]
[ 5 6 7 8]
[ 9 10 11 12]]
b: [[2 3]
[6 7]]
a_aft_changeb:
[[ 1 100 3 4]
[ 5 6 7 8]
[ 9 10 11 12]]
c_1:
[5678]
c_2:
[[5 6 7 8]]
twod_index:
[38]
threed_index
[261] d:
[[11 2 3]
[ 4 15 6]
[ 7 8 19]
[10 11 12]]
bool_idx:
[[false false]
[ true true]
[ true true]] [34
56] [33
4] x[1:7:2]:
[135] x[-2:10]:
[89]
x[-3:3:-1]:
[7654]
y[1:2]:
[[[4]
[5][6]]]
y[1]:
[[4]
[5][6]]
y[..., 0]:
[[1 2 3]
[4 5 6]]
y[:,np.newaxis,:,:]:
[[[[1]
[2][3]]]
[[[4]
[5][6]]]]
z[rows_new, cols]:
[[ 0 2]
[ 9 11]]
z[np.ix_(rows, cols)]:
[[ 0 2]
[ 9 11]]
z[1:2, 1:3]:
[[4 5]]
z[1:2, [1, 2]]:
[[4 5]]
xx[~np.isnan(xx)]:
[ 1.
2.3.]
yy: [[ 1. 99.]
[ 98. 3.]]
zz[rowsum <= 2, :]:
[[0 1]
[1 1]]
rows:
[false true false true]
xy[np.ix_(rows, cols)]:
[[ 3 5]
[ 9 11]]
rows1:
[13]
xy[rows1[:, np.newaxis], cols]:
[[ 3 5]
[ 9 11]]
numpy中常用的函式
詳細的random模組中的其他函式 np.random.rand 3,2 array 0.14022471,0.96360618 random 0.37601032,0.25528411 random 0.49313049,0.94909878 random np.random.randint 2,...
numpy中常用陣列介紹
numpy提供了高效儲存和操作密集資料快取的介面，numpy陣列幾乎是整個python資料科學工具生態系統的核心，因此學習如何有效地使用numpy是非常值得的。如果你安裝的是anaconda，那麼你已經安裝好了numpy，可以使用了。通過import numpy as np來匯入numpy 這裡as...
深度學習中常用的代價函式
1.二次代價函式 quadratic cost 其中，c表示代價函式，x表示樣本，y表示實際值，a表示輸出值，n表示樣本的總數。為簡單起見，使用乙個樣 本為例進行說明，此時二次代價函式為 假如我們使用梯度下降法 gradient descent 來調整權值引數的大小，權值w和偏置b的梯度推導如下 其...