What does unsqueeze do in pytorch? - python

lower_bounds = torch.max(set_1[:, :2].unsqueeze(1),
set_2[:, :2].unsqueeze(0)) #(n1, n2, 2)
This code snippet uses unsqueeze(1) for one tensor, but unsqeeze(0) for another. What is the difference between them?

unsqueeze turns an n-dimensionsal tensor into an n+1-dimensional one, by adding an extra dimension of zero depth. However, since it is ambiguous which axis the new dimension should lie across (i.e. in which direction it should be "unsqueezed"), this needs to be specified by the dim argument.
Hence the resulting unsqueezed tensors have the same information, but the indices used to access them are different.
Here is a visual representation of what squeeze/unsqueeze do for an effectively 2d matrix, where it is going from a 2d tensor to a 3d one, and hence there are 3 choices for the new dimension's position:

Related

Efficiently filter 3D matrix in numpy with variable 2D masks

I have a 3D numpy array points of dimensions [10000x3000x128] where the first dimension is the number of frames, the second dimension the number of points in each frame and the third dimension is a 128-element feature vector associated to each point. What I want to do is to efficiently filter the points in each frame by using a boolean 2D mask of dimensions [10000x3000] and for each of the selected points also take the related 128-dim vector of features. Moreover, in output I need still a 3D vector and not a merged 2D vector and possibly avoid any for loop.
Actually what I'm doing is:
# example of points
points = np.array([10000, 3000, 128])
# fg, bg = 2D dimensional boolean np.array
# init empty lists
fg_points, bg_points = [], []
for i in range(points.shape[0]):
fg_mask_tmp, bg_mask_tmp = fg[i], bg[i]
fg_points.append(points[i,fg_mask_tmp,:])
bg_points.append(points[i,bg_mask_tmp,:])
fg_features, bg_features = np.array(fg_points), np.array(bg_points)
But this is a quite naive solution that for sure can be improved in a more numpy-like way.
In addition, I also tried other solutions as:
fg_features = points[fg,:]
But this solution does not preserve the dimensions of the array merging the two first dimensions since the number of filtered points for each frame can vary.
Another solution I tried is to enlarge the 2D masks by appending a [128] true value to the last dimension, but with any successful result.
Dos anyone know a possible efficient solution?
Thank you in advance for any help!

Does np.dot automatically transpose vectors?

I am trying to calculate the first and second order moments for a portfolio of stocks (i.e. expected return and standard deviation).
expected_returns_annual
Out[54]:
ticker
adj_close CNP 0.091859
F -0.007358
GE 0.095399
TSLA 0.204873
WMT -0.000943
dtype: float64
type(expected_returns_annual)
Out[55]: pandas.core.series.Series
weights = np.random.random(num_assets)
weights /= np.sum(weights)
returns = np.dot(expected_returns_annual, weights)
So normally the expected return is calculated by
(x1,...,xn' * (R1,...,Rn)
with x1,...,xn are weights with a constraint that all the weights have to sum up to 1 and ' means that the vector is transposed.
Now I am wondering a bit about the numpy dot function, because
returns = np.dot(expected_returns_annual, weights)
and
returns = np.dot(expected_returns_annual, weights.T)
give the same results.
I tested also the shape of weights.T and weights.
weights.shape
Out[58]: (5,)
weights.T.shape
Out[59]: (5,)
The shape of weights.T should be (,5) and not (5,), but numpy displays them as equal (I also tried np.transpose, but there is the same result)
Does anybody know why numpy behave this way? In my opinion the np.dot product automatically shape the vector the right why so that the vector product work well. Is that correct?
Best regards
Tom
The semantics of np.dot are not great
As Dominique Paul points out, np.dot has very heterogenous behavior depending on the shapes of the inputs. Adding to the confusion, as the OP points out in his question, given that weights is a 1D array, np.array_equal(weights, weights.T) is True (array_equal tests for equality of both value and shape).
Recommendation: use np.matmul or the equivalent # instead
If you are someone just starting out with Numpy, my advice to you would be to ditch np.dot completely. Don't use it in your code at all. Instead, use np.matmul, or the equivalent operator #. The behavior of # is more predictable than that of np.dot, while still being convenient to use. For example, you would get the same dot product for the two 1D arrays you have in your code like so:
returns = expected_returns_annual # weights
You can prove to yourself that this gives the same answer as np.dot with this assert:
assert expected_returns_annual # weights == expected_returns_annual.dot(weights)
Conceptually, # handles this case by promoting the two 1D arrays to appropriate 2D arrays (though the implementation doesn't necessarily do this). For example, if you have x with shape (N,) and y with shape (M,), if you do x # y the shapes will be promoted such that:
x.shape == (1, N)
y.shape == (M, 1)
Complete behavior of matmul/#
Here's what the docs have to say about matmul/# and the shapes of inputs/outputs:
If both arguments are 2-D they are multiplied like conventional matrices.
If either argument is N-D, N > 2, it is treated as a stack of matrices residing in the last two indexes and broadcast accordingly.
If the first argument is 1-D, it is promoted to a matrix by prepending a 1 to its dimensions. After matrix multiplication the prepended 1 is removed.
If the second argument is 1-D, it is promoted to a matrix by appending a 1 to its dimensions. After matrix multiplication the appended 1 is removed.
Notes: the arguments for using # over dot
As hpaulj points out in the comments, np.array_equal(x.dot(y), x # y) for all x and y that are 1D or 2D arrays. So why do I (and why should you) prefer #? I think the best argument for using # is that it helps to improve your code in small but significant ways:
# is explicitly a matrix multiplication operator. x # y will raise an error if y is a scalar, whereas dot will make the assumption that you actually just wanted elementwise multiplication. This can potentially result in a hard-to-localize bug in which dot silently returns a garbage result (I've personally run into that one). Thus, # allows you to be explicit about your own intent for the behavior of a line of code.
Because # is an operator, it has some nice short syntax for coercing various sequence types into arrays, without having to explicitly cast them. For example, [0,1,2] # np.arange(3) is valid syntax.
To be fair, while [0,1,2].dot(arr) is obviously not valid, np.dot([0,1,2], arr) is valid (though more verbose than using #).
When you do need to extend your code to deal with many matrix multiplications instead of just one, the ND cases for # are a conceptually straightforward generalization/vectorization of the lower-D cases.
I had the same question some time ago. It seems that when one of your matrices is one dimensional, then numpy will figure out automatically what you are trying to do.
The documentation for the dot function has a more specific explanation of the logic applied:
If both a and b are 1-D arrays, it is inner product of vectors
(without complex conjugation).
If both a and b are 2-D arrays, it is matrix multiplication, but using
matmul or a # b is preferred.
If either a or b is 0-D (scalar), it is equivalent to multiply and
using numpy.multiply(a, b) or a * b is preferred.
If a is an N-D array and b is a 1-D array, it is a sum product over
the last axis of a and b.
If a is an N-D array and b is an M-D array (where M>=2), it is a sum
product over the last axis of a and the second-to-last axis of b:
In NumPy, a transpose .T reverses the order of dimensions, which means that it doesn't do anything to your one-dimensional array weights.
This is a common source of confusion for people coming from Matlab, in which one-dimensional arrays do not exist. See Transposing a NumPy Array for some earlier discussion of this.
np.dot(x,y) has complicated behavior on higher-dimensional arrays, but its behavior when it's fed two one-dimensional arrays is very simple: it takes the inner product. If we wanted to get the equivalent result as a matrix product of a row and column instead, we'd have to write something like
np.asscalar(x # y[:, np.newaxis])
adding a trailing dimension to y to turn it into a "column", multiplying, and then converting our one-element array back into a scalar. But np.dot(x,y) is much faster and more efficient, so we just use that.
Edit: actually, this was dumb on my part. You can, of course, just write matrix multiplication x # y to get equivalent behavior to np.dot for one-dimensional arrays, as tel's excellent answer points out.
The shape of weights.T should be (,5) and not (5,),
suggests some confusion over the shape attribute. shape is an ordinary Python tuple, i.e. just a set of numbers, one for each dimension of the array. That's analogous to the size of a MATLAB matrix.
(5,) is just the way of displaying a 1 element tuple. The , is required because of older Python history of using () as a simple grouping.
In [22]: tuple([5])
Out[22]: (5,)
Thus the , in (5,) does not have a special numpy meaning, and
In [23]: (,5)
File "<ipython-input-23-08574acbf5a7>", line 1
(,5)
^
SyntaxError: invalid syntax
A key difference between numpy and MATLAB is that arrays can have any number of dimensions (upto 32). MATLAB has a lower boundary of 2.
The result is that a 5 element numpy array can have shapes (5,), (1,5), (5,1), (1,5,1)`, etc.
The handling of a 1d weight array in your example is best explained the np.dot documentation. Describing it as inner product seems clear enough to me. But I'm also happy with the
sum product over the last axis of a and the second-to-last axis of b
description, adjusted for the case where b has only one axis.
(5,) with (5,n) => (n,) # 5 is the common dimension
(n,5) with (5,) => (n,)
(n,5) with (5,1) => (n,1)
In:
(x1,...,xn' * (R1,...,Rn)
are you missing a )?
(x1,...,xn)' * (R1,...,Rn)
And the * means matrix product? Not elementwise product (.* in MATLAB)? (R1,...,Rn) would have size (n,1). (x1,...,xn)' size (1,n). The product (1,1).
By the way, that raises another difference. MATLAB expands dimensions to the right (n,1,1...). numpy expands them to the left (1,1,n) (if needed by broadcasting). The initial dimensions are the outermost ones. That's not as critical a difference as the lower size 2 boundary, but shouldn't be ignored.

What does (n,) mean in the context of numpy and vectors?

I've tried searching StackOverflow, googling, and even using symbolhound to do character searches, but was unable to find an answer. Specifically, I'm confused about Ch. 1 of Nielsen's Neural Networks and Deep Learning, where he says "It is assumed that the input a is an (n, 1) Numpy ndarray, not a (n,) vector."
At first I thought (n,) referred to the orientation of the array - so it might refer to a one-column vector as opposed to a vector with only one row. But then I don't see why we need (n,) and (n, 1) both - they seem to say the same thing. I know I'm misunderstanding something but am unsure.
For reference a refers to a vector of activations that will be input to a given layer of a neural network, before being transformed by the weights and biases to produce the output vector of activations for the next layer.
EDIT: This question equivocates between a "one-column vector" (there's no such thing) and a "one-column matrix" (does actually exist). Same for "one-row vector" and "one-row matrix".
A vector is only a list of numbers, or (equivalently) a list of scalar transformations on the basis vectors of a vector space. A vector might look like a matrix when we write it out, if it only has one row (or one column). Confusingly, we will sometimes refer to a "vector of activations" but actually mean "a single-row matrix of activation values transposed so that it is a single-column."
Be aware that in neither case are we discussing a one-dimensional vector, which would be a vector defined by only one number (unless, trivially, n==1, in which case the concept of a "column" or "row" distinction would be meaningless).
In numpy an array can have a number of different dimensions, 0, 1, 2 etc.
The typical 2d array has dimension (n,m) (this is a Python tuple). We tend to describe this as having n rows, m columns. So a (n,1) array has just 1 column, and a (1,m) has 1 row.
But because an array may have just 1 dimension, it is possible to have a shape (n,) (Python notation for a 1 element tuple: see here for more).
For many purposes (n,), (1,n), (n,1) arrays are equivalent (also (1,n,1,1) (4d)). They all have n terms, and can be reshaped to each other.
But sometimes that extra 1 dimension matters. A (1,m) array can multiply a (n,1) array to produce a (n,m) array. A (n,1) array can be indexed like a (n,m), with 2 indices, x[:,0] where as a (n,) only accepts x[0].
MATLAB matrices are always 2d (or higher). So people transfering ideas from MATLAB tend to expect 2 dimensions. There is a np.matrix subclass that supposed to imitate that.
For numpy programmers the distinctions between vector, row vector, column vector, matrix are loose and relatively unimportant. Or the use is derived from the application rather than from numpy itself. I think that's what's happening with this network book - the notation and expectations come from outside of numpy.
See as well this answer for how to interpret the shapes with respect to the data stored in ndarrays. It also provides insight on how to use .reshape: https://stackoverflow.com/a/22074424/3277902
(n,) is a tuple of length 1, whose only element is n. (The syntax isn't (n) because that's just n instead of making a tuple.)
If an array has shape (n,), that means it's a 1-dimensional array with a length of n along its only dimension. It's not a row vector or a column vector; it doesn't have rows or columns. It's just a vector.

Decomposing 3rd Order Tensor in Python

I have a tensor in the shape (n_samples, n_steps, n_features). I want to decompose this into a tensor of shape (n_samples, n_components).
I need a method of decomposition that has a .fit(...) so that I can apply the same decomposition to a new batch of samples. I have been looking at Tucker Decomposition and PARAFAC Decomposition, but neither have that crucial .fit(...) and .transform(...) functionality. (Or at least I think they don't?)
I could use PCA and train it on a representative sample and then call .transform(...) on the remaining samples, but I would rather have some sort of tensor decomposition that can handle all of the samples at once, so as to get a better idea of the differences between each sample.
This is what I mean by "tensor":
In fact tensors are merely a generalisation of scalars and vectors; a scalar is a zero rank tensor, and a vector is a first rank tensor. The rank (or order) of a tensor is defined by the number of directions (and hence the dimensionality of the array) required to describe it.
If you have any questions, please ask, I'll try to clarify my problem if needed.
EDIT: The best solution would be some type of kernel but I have yet to find a kernel that can deal with n-rank Tensors and not just 2D data
You can do this using the development (master) version of TensorLy. Specifically, you can use the new partial_tucker function (it is not yet updated in the documentation...).
Note that the following solution preserves the structure of the tensor, i.e. a tensor of shape (n_samples, n_steps, n_features) is decomposed into a (smaller) tensor of shape (n_samples, n_components_1, n_components_2).
Code
Short answer: this is a very basic class that does what you want (and it would work on tensors of arbitrary order).
import tensorly as tl
from tensorly.decomposition._tucker import partial_tucker
class TensorPCA:
def __init__(self, ranks, modes):
self.ranks = ranks
self.modes = modes
def fit(self, tensor):
self.core, self.factors = partial_tucker(tensor, modes=self.modes, ranks=self.ranks)
return self
def transform(self, tensor):
return tl.tenalg.multi_mode_dot(tensor, self.factors, modes=self.modes, transpose=True)
Usage
Given an input tensor, you can use the previous class by first instantiating it with the desired ranks (size of the core tensor) and modes on which to perform the decomposition (in your 3D case, 1 and 2 since indexing starts at zero):
tpca = TensorPCA(ranks=[4, 5], modes=[1, 2])
tpca.fit(tensor)
Given a new tensor originally called new_tensor, you can project it using the transform method:
tpca.transform(new_tensor)
Explanation
Let's go through the code with an example: first let's import the necessary bits:
import numpy as np
import tensorly as tl
from tensorly.decomposition._tucker import partial_tucker
We then generate a random tensor:
tensor = np.random.random((10, 11, 12))
The next step is to decompose it along its second and third dimensions, or modes (as the first dimension corresponds to the samples):
core, factors = partial_tucker(tensor, modes=[1, 2], ranks=[4, 5])
The core corresponds to the transformed input tensor while factors is a list of two projection matrices, one for the second mode and one for the third mode. Given a new tensor, you can project it to the same subspace (the transform method) by projecting each of its last two dimensions:
tl.tenalg.multi_mode_dot(tensor, factors, modes=[1, 2], transpose=True)
The transposition here is equivalent to an inverse since the factors are orthogonal.
Finally, a note on the terminology: in general, even though it is sometimes done, it is probably best to not use interchangeably order and rank of a tensor. The order of a tensor is simply its number of dimensions while the rank of a tensor is usually a much more complicated notion which you could think of as a generalization of the notion of matrix rank.

Axis elimination

I'm having a trouble understanding the concept of Axis elimination in numpy. Suppose I have the following 2D matrix:
A =
1 2 3
3 4 5
6 7 8
Ok I understand that sum(A, axis=0) will sum each column down and will give a 1D array with 3 elements. I also understand that sum(A, axis=1) will sum each row.
But my trouble is when I read that axis=0 eliminates the 0th axis and axis=1 eliminates the 1th axis. Also sometime people mention "reduce" instead of "eliminate". I'm unable to understand what does that eliminate. For example sum(A, axis=0) will sum each column from top to bottom, but I don't see elimination or reduction here. What's the point? The same also for sum(A,axis=1).
AND how is it for higher dimensions?
p.s. I always confused between matrix dimensions and array dimensions. I wished that people who write the numpy documentation makes this distinction very clear.
http://docs.scipy.org/doc/numpy/reference/generated/numpy.ufunc.reduce.html
Reduces a‘s dimension by one, by applying ufunc along one axis.
For example, add.reduce() is equivalent to sum().
In numpy, the base class is ndarray - a multidimensional array (can 0d, 1d, or more)
http://docs.scipy.org/doc/numpy/reference/arrays.ndarray.html
Matrix is a subclass of array
http://docs.scipy.org/doc/numpy/reference/arrays.classes.html
Matrix objects are always two-dimensional
The history of the numpy Matrix is old, but basically it's meant to resemble the MATLAB matrix object. In the original MATLAB nearly everything was a matrix, which was always 2d. Later they generalized it to allow more dimensions. But it can't have fewer dimensions. MATLAB does have 'vectors', but they are just matrices with one dimension being 1 (row vector versus column vector).
'axis elimination' is not a common term when working with numpy. It could, conceivably, refer to any of several ways that reduce the number of dimensions of an array. Reduction, as in sum(), is one. Indexing is another: a[:,0,:]. Reshaping can also change the number of dimensions. np.squeeze is another.

Categories