Skip to content

Latest commit

 

History

History
508 lines (360 loc) · 10.2 KB

exercises-1-20.livemd

File metadata and controls

508 lines (360 loc) · 10.2 KB
Raw

Exercises: 1-20

Mix.install([{:nx, "~> 0.6"}])

Introduction

Inspired by the Python notebook 100 Numpy Exercises.

1-10

1. Install Nx in a Livebook. (★☆☆)

# Add your solution here.
Example solution 
Mix.install([{:nx, "~> 0.6"}])

2. Create a 1-D tensor of 10 zeros. (★☆☆)

# Add your solution here.
Example solution 
Nx.broadcast(0, {10})

3. Find the number of elements in tensor. (★☆☆)

tensor = Nx.tensor([[1, 2, 3], [4, 5, 6]])
# Add your solution here.
Example solution 
Nx.size(tensor)

4. Find the number of bytes of memory in tensor. (★☆☆)

tensor = Nx.tensor([[1, 2, 3], [4, 5, 6]])
# Add your solution here.
Example solution 
Nx.byte_size(tensor)

5a. Use Nx.sum/2 to find the sum of all elements of tensor. (★☆☆)

tensor = Nx.tensor([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
# Add your solution here.
Example solution 
Nx.sum(tensor)

5b. Read the documentation for Nx.sum/2 then provide the correct option to sum across the rows. (★☆☆)

# Add your solution here.
Example solution 
Nx.sum(tensor, axes: [1])
Tip: You can also hover over a function inside Livebook code cells to display its documentation.

6. Create a tensor of zeros of size 10 but where the fifth value is 1. (★☆☆)

# Add your solution here.
Example solution 
zeros = Nx.broadcast(0, {10})
index = Nx.tensor([4])
Nx.indexed_put(zeros, index, 1)

7. Create a 3x3 tensor with values ranging from 0 to 8. (★☆☆)

# Add your solution here.
Example solution 
Nx.iota({3, 3})

8. Create a tensor with values ranging from 10 to 49. (★☆☆)

# Add your solution here.
Example solution 1 
Nx.iota({39})
|> Nx.add(10)
Example solution 2 
Nx.linspace(10, 49, n: 39, type: :s64)

9. Reverse tensor (first element becomes last). (★☆☆)

tensor = Nx.tensor([2, 4, 6, 8])
# Add your solution here.
Example solution 
Nx.reverse(tensor)

10a. Given an initial tensor, build a "mask" of non-zero elements. That is, build a second tensor with the same shape as the original, but that has a 1 wherever the original has a non-zero element and a 0 elsewhere. (★☆☆)

tensor = Nx.tensor([1, 2, 0, 0, 4, 0])
# Add your solution here.
Example solution 
mask = Nx.not_equal(tensor, 0)

10b. Use the mask from 10a to replace each 0 from tensor with -1. (★☆☆)

# Add your solution here.
Example solution 
Nx.select(mask, tensor, -1)

11-20

11. Create a 3x3 identity tensor. (★☆☆)

# Add your solution here.
Example solution 
Nx.eye(3)

12. Create a 3x3x3 tensor with random values. (★☆☆)

# Add your solution here.
Example solution 
key = Nx.Random.key(0)
{random, _} = Nx.Random.normal(key, shape: {3, 3, 3})
random

13. Create a random 10x10 tensor then find its minimum and maximum values. (★☆☆)

# Add your solution here.
Example solution 
key = Nx.Random.key(0)
{tensor, _} = Nx.Random.normal(key, shape: {10, 10})

%{
  min: Nx.reduce_min(tensor),
  max: Nx.reduce_max(tensor)
}

14. Create a random 1D tensor of size 30 then find its mean. (★☆☆)

# Add your solution here.
Example solution 
key = Nx.Random.key(0)
{tensor, _} = Nx.Random.normal(key, shape: {30})

Nx.mean(tensor)

15. Create a 4x4 tensor with 1 on the border and 0 inside. (★☆☆)

# Add your solution here.
Example solution 
Nx.broadcast(1, {4, 4})
|> Nx.put_slice([1, 1], Nx.broadcast(0, {2, 2}))

16. Add a border of 0 around tensor (end result will be a 5x5 tensor). (★☆☆)

tensor = Nx.broadcast(1, {3, 3})
# Add your solution here.
Example solution 
Nx.pad(tensor, 0, [{1, 1, 0}, {1, 1, 0}])

17. Determine the results of the following expressions. (★☆☆)

nan = Nx.Constants.nan()
Nx.multiply(0, nan)
Nx.equal(nan, nan)
Nx.greater(nan, nan)
Nx.subtract(nan, nan)

# Add your solution here.
Example solution 
#Nx.Tensor<
  f32
  NaN
>
#Nx.Tensor<
  u8
  0
>
#Nx.Tensor<
  u8
  0
>
#Nx.Tensor<
  f32
  NaN
>

18. Create a 5x5 tensor with values 1,2,3,4 just below the diagonal. (★☆☆)

# Add your solution here.
Example solution 
Nx.tensor([1, 2, 3, 4])
|> Nx.make_diagonal(offset: -1)

19. Create a 8x8 tensor of 0 and 1 in a checkerboard pattern with 0 as the first element using Nx.tile. (★☆☆)

# Add your solution here.
Example solution 
Nx.tensor([[0, 1], [1, 0]])
|> Nx.tile([4, 4])

20. Produce the same checkerboard pattern from exercise 19, but without using Nx.tile. (★☆☆)

# Add your solution here.
# Hint: try using `Nx.iota` in combination with `Nx.remainder`.
Example solution 1 
t = Nx.iota({8, 1})

Nx.transpose(t)
|> Nx.add(t)
|> Nx.remainder(2)
Example solution 2 
Nx.iota({9, 9})
|> Nx.remainder(2)
|> Nx.slice([0, 0], [8, 8])