lovely-numpy 0.2.0

💟 Lovely numpy

💟 Lovely NumPy

Read full docs | ❤️ Lovely Tensors | 💘 Lovely JAX | Discord

Install

pip install lovely-numpy

How to use

How often do you find yourself debugging NumPy code? You dump your variable to the cell output, and see this:

numbers

array([[[-0.3541, -0.1975, -0.6715],
        [-0.3369, -0.1975, -0.9853],
        ...,
        [-0.4739, -0.3725, -0.689 ],
        [ 2.2489,  2.4111,  2.396 ]],

       [[-0.4054, -0.25  , -0.7238],
        [-0.4226, -0.2325, -1.0724],
        ...,
        [-0.8507, -0.6702, -1.0201],
        [ 2.1633,  2.3585,  2.3263]],

       ...,

       [[-0.8507, -0.3901, -1.1944],
        [-0.7822, -0.2325, -1.4559],
        ...,
        [-1.5014, -1.2304, -1.4733],
        [ 2.1804,  2.4111,  2.4308]],

       [[-0.8335, -0.4076, -1.2293],
        [-0.8164, -0.285 , -1.5256],
        ...,
        [-1.5528, -1.2829, -1.5256],
        [ 2.1119,  2.341 ,  2.3611]]], dtype=float32)

Was it really useful for you, as a human, to see all these numbers?

What is the shape? The size?
What are the statistics?
Are any of the values nan or inf?
Is it an image of a man holding a tench?

from lovely_numpy import Lo

Lo and behold!

Lo(numbers)

array[196, 196, 3] f32 n=115248 x∈[-2.118, 2.640] μ=-0.388 σ=1.073

Better, eh?

Lo(numbers[1,:6,1]) # Still shows values if there are not too many.

array[6] f32 x∈[-0.408, -0.232] μ=-0.340 σ=0.075 [-0.250, -0.232, -0.338, -0.408, -0.408, -0.408]

spicy = numbers[0,:12,0].copy()

spicy[0] *= 10000
spicy[1] /= 10000
spicy[2] = float('inf')
spicy[3] = float('-inf')
spicy[4] = float('nan')

spicy = spicy.reshape((2,6))
Lo(spicy) # Spicy stuff

array[2, 6] f32 n=12 x∈[-3.541e+03, -3.369e-05] μ=-393.776 σ=1.113e+03 +Inf! -Inf! NaN!

Lo(np.zeros((10, 10))) # A zero array - make it obvious

array[10, 10] all_zeros

Lo(spicy, verbose=True)

array[2, 6] f32 n=12 x∈[-3.541e+03, -3.369e-05] μ=-393.776 σ=1.113e+03 +Inf! -Inf! NaN!
array([[-3540.5432,    -0.    , ...,        nan,    -0.4054],
       [   -0.4226,    -0.4911, ...,    -0.5424,    -0.5082]],
      dtype=float32)

Going .deeper

Lo(numbers.transpose(2,1,0), depth=1)

array[3, 196, 196] f32 n=115248 x∈[-2.118, 2.640] μ=-0.388 σ=1.073
  array[196, 196] f32 n=38416 x∈[-2.118, 2.249] μ=-0.324 σ=1.036
  array[196, 196] f32 n=38416 x∈[-1.966, 2.429] μ=-0.274 σ=0.973
  array[196, 196] f32 n=38416 x∈[-1.804, 2.640] μ=-0.567 σ=1.178

# You can go deeper if you need to
Lo(numbers[:3,:3,:5], depth=2)

array[3, 3, 3] f32 n=27 x∈[-1.125, -0.197] μ=-0.544 σ=0.291
  array[3, 3] f32 n=9 x∈[-0.985, -0.197] μ=-0.481 σ=0.276 [[-0.354, -0.197, -0.672], [-0.337, -0.197, -0.985], [-0.405, -0.303, -0.881]]
    array[3] f32 x∈[-0.672, -0.197] μ=-0.408 σ=0.197 [-0.354, -0.197, -0.672]
    array[3] f32 x∈[-0.985, -0.197] μ=-0.507 σ=0.343 [-0.337, -0.197, -0.985]
    array[3] f32 x∈[-0.881, -0.303] μ=-0.530 σ=0.252 [-0.405, -0.303, -0.881]
  array[3, 3] f32 n=9 x∈[-1.072, -0.232] μ=-0.545 σ=0.289 [[-0.405, -0.250, -0.724], [-0.423, -0.232, -1.072], [-0.491, -0.338, -0.968]]
    array[3] f32 x∈[-0.724, -0.250] μ=-0.460 σ=0.197 [-0.405, -0.250, -0.724]
    array[3] f32 x∈[-1.072, -0.232] μ=-0.576 σ=0.360 [-0.423, -0.232, -1.072]
    array[3] f32 x∈[-0.968, -0.338] μ=-0.599 σ=0.268 [-0.491, -0.338, -0.968]
  array[3, 3] f32 n=9 x∈[-1.125, -0.285] μ=-0.605 σ=0.293 [[-0.474, -0.303, -0.828], [-0.474, -0.285, -1.125], [-0.542, -0.390, -1.020]]
    array[3] f32 x∈[-0.828, -0.303] μ=-0.535 σ=0.219 [-0.474, -0.303, -0.828]
    array[3] f32 x∈[-1.125, -0.285] μ=-0.628 σ=0.360 [-0.474, -0.285, -1.125]
    array[3] f32 x∈[-1.020, -0.390] μ=-0.651 σ=0.268 [-0.542, -0.390, -1.020]

Now in .rgb color

The important queston - is it our man?

Lo(numbers).rgb

Maaaaybe? Looks like someone normalized him.

in_stats = ( (0.485, 0.456, 0.406),     # mean 
             (0.229, 0.224, 0.225) )    # std

# numbers.rgb(in_stats, cl=True) # For channel-last input format
Lo(numbers).rgb(denorm=in_stats)

It’s indeed our hero, the Tenchman!

See the .chans

# .chans will map values betwen [0,1] to colors.
# Make our values fit into that range to avoid clipping.
mean = np.array(in_stats[0])
std = np.array(in_stats[1])
numbers_01 = (numbers*std + mean)
Lo(numbers_01)

array[196, 196, 3] n=115248 x∈[-4.053e-09, 1.000] μ=0.361 σ=0.248

Lo(numbers_01).chans

Grouping

# Make 8 images with progressively higher brightness and stack them 2x2x2.
eight_images = (np.stack([numbers]*8) + np.linspace(-2, 2, 8)[:,None,None,None])
eight_images = (eight_images
                     *np.array(in_stats[1])
                     +np.array(in_stats[0])
                ).clip(0,1).reshape(2,2,2,196,196,3)
            
Lo(eight_images)

array[2, 2, 2, 196, 196, 3] n=921984 x∈[0., 1.000] μ=0.382 σ=0.319

Lo(eight_images).rgb

Without Lo

from lovely_numpy import lovely, rgb, chans

lovely(numbers) # Returns `str`. `Lo(x)` returns a wrapper object with a `__repr__` and other methods.

'array[196, 196, 3] f32 n=115248 x∈[-2.118, 2.640] μ=-0.388 σ=1.073'

rgb(numbers, denorm=in_stats) # Returns a `PIL.Image.Image`, just like Lo(x).rgb

chans(numbers*0.3+0.5) # Also a `PIL.Image.Image`