Fixes to questions 2, 9, 18, 31, 35, 44, 52, 56, 66 & 78 #106
Conversation
* Treating `I` as `np.ubyte` causes multiplication with 256 to overflow. * There is no need to apply `np.unique` to `I.
|
Thanks. There's a new format where only file has to be changed. Could you use it ? Also, it would be better to have one PR per proposed changed (would make discussion easier). |
|
I only updated the |
| @@ -72,8 +72,8 @@ print(Z) | |||
| `hint: reshape` | |||
|
|
|||
| ```python | |||
| nz = np.nonzero([1,2,0,0,4,0]) | |||
| print(nz) | |||
| Z = np.arange(9).reshape((3, 3)) | |||
There was a problem hiding this comment.
The current solution corresponds to question 10.
There was a problem hiding this comment.
My bad... Thanks for fixing it!
| B = np.ones(3)*2 | ||
| C = np.ones(3)*3 | ||
| A = np.ones(3) | ||
| B = np.full(3, 2) |
There was a problem hiding this comment.
I prefer the version with ones (in order to keep it simple)
There was a problem hiding this comment.
Okay, I will rebase and drop commit. Should the C matrix stay as well?
|
|
||
| ```python | ||
| Z = np.random.random((10,2)) | ||
| X,Y = Z[:,0], Z[:,1] | ||
| R = np.sqrt(X**2+Y**2) | ||
| R = np.hypot(Y,X) |
There was a problem hiding this comment.
The order is such so that it's consistent with the order of arguments in the arctan2 invocation. It's my personal mnemonic, but I will change it if you insist.
| @@ -673,10 +673,10 @@ print(F) | |||
| # Author: Nadav Horesh | |||
|
|
|||
| w,h = 16,16 | |||
| I = np.random.randint(0,2,(h,w,3)).astype(np.ubyte) | |||
| I = np.random.randint(0,2,(h,w,3)) | |||
There was a problem hiding this comment.
Why did you remove the cast to ubyte ?
There was a problem hiding this comment.
In a sufficiently long array, the expected number of different "colors" should be 2**3 == 8. For the solution to work, F must be a matrix of at least 24-bit integers, but this is not what happens on my machine: I[..., 1] * 256 has type np.uint16, and the same happens with I[..., 0] * 256 * 256, leading to overflow. Writing (I[..., 0] * 256) * 256 doesn't solve the problem. A cast to a sufficient wide type has to happen at some point, or the initial cast has to go.
>>> import numpy as np
>>> w, h = 16, 16
>>> I = np.random.randint(0, 2, (h, w, 3))
>>> I2 = I.astype(np.ubyte)
>>> F = I[...,0]*256*256 + I[...,1]*256 +I[...,2]
>>> F2 = I2[...,0]*256*256 + I2[...,1]*256 +I2[...,2]
>>> print(len(np.unique(F)), len(np.unique(F2)))
8 4
>>> print(F.dtype, F2.dtype)
int64 uint16
| @@ -826,9 +826,9 @@ np.negative(Z, out=Z) | |||
| def distance(P0, P1, p): | |||
| T = P1 - P0 | |||
| L = (T**2).sum(axis=1) | |||
| U = -((P0[:,0]-p[...,0])*T[:,0] + (P0[:,1]-p[...,1])*T[:,1]) / L | |||
| U = ((P0 - p)*T).sum(axis=1) / L | |||
Fixes #98
Fixes #101
I have added a few more fixes and simplifications, let me know what you think.
I ran all the generators; some of the output is due to older commits.