Please remove the space after : on this and similar lines.

I don't particularly like the m_ptr1/2 nomenclature. Perhaps m_start, m_stop?

Just in case someone passes in a mutable (for instance a numpy int instead of a Python int), I would do 'm = m + X.shape[0]' here to avoid side effects.

Just out of curiousity (and for learning), could you please tell me how would the behaviour of the code that I've written, change for a numpy int ?

I don't get that:

>>> x = y = np.int(0)
>>> y += 1
>>> x, y
(0, 1)

Gael, where you thinking of 0-d arrays?

>>> x = y = array(0)
>>> x += 1
>>> x, y
(array(1), array(1))

I'm not sure if guarding for this is worthwhile. It's very likely that a later refactoring round undoes this; callers should take care not to produce these values.

Yes I was. 

Maybe I am paranoid but each time I see an in place modification I check that it does not apply to input arguments. Thus I think that l might see the potential problem also in a later code review. 

I think that it is good practice not to let such a code.  If m is an intégré there is no computational benefit to doing the +=.

@@ -56,3 +57,109 @@ def inplace_column_scale(X, scale):
else:
raise TypeError(
"Unsupported type; expected a CSR or CSC sparse matrix.")
+
+
+def swap_row_csc(X, m, n):

• """
• Swaps two rows of a CSC matrix in-place.

• Parameters

---

• X : scipy.sparse.csc_matrix, shape=(n_samples, n_features)
• m : int, index of first sample
• n : int, index of second sample
• """
• if m < 0:

•    m += X.shape[0]
  

  I don't get that:

x = y = np.int(0)
y += 1
x, y
(0, 1)
Gael, where you thinking of 0-d arrays?

x = y = array(0)
x += 1
x, y
(array(1), array(1))
I'm not sure if guarding for this is worthwhile. It's very likely that a later refactoring round undoes this; callers should take care not to produce these values.

—
Reply to this email directly or view it on GitHub.

It's a very Pythonic shorthand, though. I think it would be cleaner to typecheck and raise a TypeError for a 0-d array:

>>> isinstance(np.array(1), (np.integer, Integral))
False

@GaelVaroquaux Is it just 0-D numpy arrays, or are there other cases also? Then it might be better to do m = m + X.shape[0]

It's a very Pythonic shorthand, though. I think it would be cleaner to
typecheck and raise a TypeError for a 0-d array:

isinstance(np.array(1), (np.integer, Integral))
False

If you whish.

Coming late to the party but it would be nice to report the actual type of sparse matrix with X.getformat() in the error message.