I have a numpy array which is 300 rows and 5 columns
X[X[:,0]==1,[1,2]]
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-154-de5e74bc1a0b> in <module>()
----> 1 X[X[:,0]==1,[1,2]]
ValueError: shape mismatch: objects cannot be broadcast to a single shape
How to make this work? I want to filter the first column (equals 1) but returns a copy of second and third column.
It's simplest to separate the two indexing cases. Look at column one first, select the appropriate rows, and then pick out columns 1 and 2 from that array:
>>> a = np.random.randint(0, 2, (3, 5))
>>> a
array([[0, 0, 0, 0, 1],
[1, 0, 1, 0, 1],
[0, 0, 1, 1, 0]])
>>> a[a[:,0] == 1][:,[1,2]]
array([[0, 1]])
The code in the question, a[a[:,0] == 1, [1,2]], look syntactically similar but is doing something different.
For example a[a[:,0] == 0, [1,2]] is, in the case of a, equivalent to a[[0,2], [1,2]]. This indexing picks out exactly two elements from a: the element at row 0, column 1 and the element at row 2, column 2. It does not select the rows and columns of the array.
a[:,0] == 1 you could write np.where(a[:,0]==1)[0] but this is less concise. Your approach looks syntactically very similar to mine but it is actually doing something different. For example, with my array a, a[a[:,0]==0, [1,2]] is equivalent to a[[0,2], [1,2]] which picks two elements out of the array: the element at (0, 1) and the element at (2, 2).