Numpy
in Study / Computer science on Data structure
Python Tool
- Tools for “ndarray” data
- ndarray: stores identical type of data
array
- Iterator like list -> ndarray
- Its dimension can be 1, 2, 3, …
import numpy as np
arr1 = np.array([3, 5, 7]) # 1D
arr2 = np.array([[3, 5, 7]]) # 2D
arr3 = np.array([[3, 5, 7], [5, 8, 9]]) # 3D
astype
- The type of ndarray’s element can change
import numpy as np
arr1 = np.array([3, 5, 7])
arr2 = arr1.astype('float64')
arr3 = arr2.astype('int32')
arange
- We can easily make ndarray with integer numbers
import numpy as np
arr1 = np.arange(1, 20) # [1, 2, ..., 19]
zeros & ones
- We can easily make ndarray with the shape that we want to make (zeros: all values are 0, ones: all values are 1)
- “dtype=”: Default-‘float64’
import numpy as np
arr1 = np.zeros((3, 4)) # 3x4 Array with 0
arr2 = np.ones((3, 4)) # 3x4 Array with 1
reshape
- We can change the shape of ndarray
- The shape should be compatible with the number of elements in ndarray
- (n, m): n x m = Number of elements in ndarray
- reshape(n, m)
- 3D: reshape(n, m, k)
- change the shape to (n, m)
- reshape(-1, m)
- It automatically finds the n-value based on the number of elements in ndarray -> 2D Array regardless of original’s dimension
- reshape(n, -1)
- It automatically finds the m-value based on the number of elements in ndarray -> 2D Array regardless of original’s dimension
- reshape(-1, 1) (used frequently)
- 2D Array regardless of original’s dimension with only one column
import numpy as np
array = np.arange(1, 17)
arr2 = array.reshape(2, 2, 2, 2)
arr3 = arr2.reshape(-1, 2)
print(array)
print(arr2)
print(arr3)
[ 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]
[[[[ 1 2]
[ 3 4]]
[[ 5 6]
[ 7 8]]]
[[[ 9 10]
[11 12]]
[[13 14]
[15 16]]]]
[[ 1 2]
[ 3 4]
[ 5 6]
[ 7 8]
[ 9 10]
[11 12]
[13 14]
[15 16]]
Indexing
- One element
arr1 = np.array([3, 5, 7])
arr2 = np.array([[3, 5, 7], [5, 8, 9]])
print(arr1[1]) # 5
print(arr2[1, 2]) # 9
- Slicing
arr1 = np.array([3, 5, 7])
arr2 = np.array([[3, 5, 7], [1, 8, 9], [10, 4, 6]])
print(arr1[:2]) # [3, 5] 1D
print(arr2[1:, 1:]) # [[8, 9], [4, 6]] 2D
print(arr2[1:, 0]) # [1, 10] 1D
- Fancy Indexing
- When its dimension is 2D: If one is fancy indexing, the other one should not be fancy indexing
- Other ndarray having the index as its data can be the part of fancy indexing (not only list)
arr1 = np.array([3, 5, 7])
arr2 = np.array([[3, 5, 7], [1, 8, 9], [10, 4, 6]])
print(arr1[[1, 2]]) # [5, 7]
print(arr2[[1, 2], 1]) # [8, 4]
print(arr2[[1, 2], 1:]) # [[8, 9], [4, 6]]
print(arr2[[1, 2]]) # [[1, 8, 9], [10, 4, 6]]
- Boolean Indexing
- Boolean Expression involving ndarray == ndarray consisting of True/False with same index of target ndarray
arr1 = np.array([3, 5, 7])
index = arr1 > 3 # [False, True, True]
arr2 = arr1[index] # [5, 7]
sort
- np.sort(): no side effect (just return)
- ndarray.sort(): side effect (return None)
- np.sort()[::-1]: descending order
- For 2D array
- axis = 0: sort between the elements that are same column
- axis = 1: sort between the elements that are same row
arr1 = np.array([5, 3, 7])
arr2 = np.array([[3, 11, 7], [1, 8, 5], [10, 9, 6]])
arr3 = np.sort(arr1)[::-1] # [7, 5, 3]
arr1.sort() # [3, 5, 7]
arr4 = np.sort(arr2, axis=0) # [[1, 8, 5], [3, 9, 6], [10, 11, 7]]
arr5 = np.sort(arr2, axis=1) # [[3, 7, 11], [1, 5, 8], [6, 9, 10]]
argsort
- It returns ndarray of index having original ndarray’s index as the elements and this ndarray’s index represents the location of these elements when it is sorted
arr1 = np.array([5, 3, 7, 6])
arr2 = np.array(["cho", "samoh", "sanford", "rivera"])
sort_index = np.argsort(arr1) # [1, 0, 3, 2] ndarray
arr3 = arr2[sort_index] # arr2 can be sorted as well by fancy indexing
# ["samoh", "cho", "rivera", "sanford"]
dot
- We can implement the multiplication between 2 matrices
arr1 = np.array([[1, 4, 3], [5, 7, 6]])
arr2 = np.array([[1, 2], [3, 1], [6, 1]])
arr3 = np.dot(arr1, arr2) # [[31, 9], [62, 23]]
transpose
- We can make transpose matrix
arr1 = np.array([[1, 4, 3], [5, 7, 6]])
arr2 = np.transpose(arr1) # [[1, 5], [4, 7], [3, 6]]