Indexing & Sorting

Indexing

Nonzero

Tensor *Tensor::nonzero(bool sort_indices = false)

Returns a tensor containing the indices of nonzero elements.

Parameters: sort_indices – Whether to sort the indices or not. (default: not sorted)
Returns: A tensor containing the indices of the nonzero elements.

Example:

Tensor* t1 = new Tensor({0,10,20,0,50,0}, {2, 3});
// [
// [0.00 10.00 20.00]
// [0.00 50.00 0.00]
// ]

Tensor* t2 = t1->nonzero();
// [1 2 4]

Where

static Tensor *Tensor::where(Tensor *condition, Tensor *A, Tensor *B)

Depending on condition, returns a tensor whith elements from A or B.

Parameters

condition – Tensor with the condition to be accomplished.
A – Input tensor.
B – Input tensor.

Returns

A tensor with the same shape with elements from A if condition holds and from B otherwise..

Example:

Tensor* t1 = Tensor::randn({2, 3});
// [
// [-0.44 -0.22 0.12]
// [1.37 0.50 -0.72]
// ]

Tensor* t2 = Tensor::zeros({2, 3});
// [
// [0.00 0.00 0.00]
// [0.00 0.00 0.00]
// ]

Tensor* condition = t1->greater(t2);
// [
// [0 0 1]
// [1 1 0]
// ]

Tensor* t3 = Tensor::where(condition, t1, t2);
// [
// [0.00 0.00 0.12]
// [1.37 0.50 0.00]
// ]

Select

Tensor *Tensor::select(const vector<string> &indices)

Returns an array with the selected indices of the tensor.

Parameters: indices – Vector of strings representing the indices to be selected. These indices must follow a Python-like syntax. Some examples: "0" , ":5" , ":" , "3:6".
Returns: Tensor

Example:

Tensor* t1 = Tensor::randn({3, 4});
// [
// [2.30 -1.13 -0.61 0.04]
// [-0.95 0.96 -0.32 -0.61]
// [0.29 -0.14 -0.42 -0.04]
// ]

Tensor* t2 = t1->select({"0", "2"});  // Number at (0, 2)
// [
// [-0.61]
// ]

Tensor* t3 = t1->select({":", "2"}); // 3rd column
// [
// [-0.61]
// [-0.32]
// [-0.42]
// ]

Tensor* t4 = t1->select({"0:2", "-2:"}); // 3rd and 4th column from 1st and 2nd row
// [
// [-0.61 0.04]
// [-0.32 -0.61]
// ]

Set Select

void Tensor::set_select(const vector<string> &indices, float value)

Sets the elements in the array using the selected indices. The indices must be specified as a vector of strings ({“0”, “:5”, “:”, “3:6”}).

Parameters

indices – Vector of strings representing the indices to be selected. These indices must follow a Python-like syntax. Some examples: "0", ":5", ":", "3:6".
value –

Returns

void

Example:

Tensor* t1 = Tensor::ones({3, 4});
t1->set_select({"0", "2"}, 2.0f); // Set element (0, 2) to "2"
// [
// [1.00 1.00 2.00 1.00]
// [1.00 1.00 1.00 1.00]
// [1.00 1.00 1.00 1.00]
// ]

Tensor* t2 = Tensor::ones({3, 4});
t2->set_select({":", "2"}, 5.0f);  // Set 2nd column to "5"
// [
// [1.00 1.00 5.00 1.00]
// [1.00 1.00 5.00 1.00]
// [1.00 1.00 5.00 1.00]
// ]

Tensor* t3 = Tensor::ones({3, 4});
t3->set_select({"0:2", "-2:"}, 7.0f);  // Set 3rd and 4th column from 1st and 2nd row to "7"
// [
// [1.00 1.00 7.00 7.00]
// [1.00 1.00 7.00 7.00]
// [1.00 1.00 1.00 1.00]
// ]

void Tensor::set_select(const vector<string> &indices, Tensor *A)

Sets the elements in the array using the selected indices. The indices must be specified as a vector of strings ({“0”, “:5”, “:”, “3:6”}).

Parameters

indices – Vector of strings representing the indices to be selected. These indices must follow a Python-like syntax. Some examples: "0", ":5", ":", "3:6".
A –

Returns

void

Example:

Tensor* t1 = Tensor::ones({3, 4});
Tensor* t2 = Tensor::full({1, 1}, 5.0f);
t1->set_select({"0", "2"}, t2); // Set element (0, 2) to "5"
// [
// [1.00 1.00 5.00 1.00]
// [1.00 1.00 1.00 1.00]
// [1.00 1.00 1.00 1.00]
// ]

Tensor* t3 = Tensor::ones({3, 4});
Tensor* t4 = Tensor::full({3, 1}, 5.0f);
t3->set_select({":", "2"}, t4);  // Set 2nd column to "5"
// [
// [1.00 1.00 5.00 1.00]
// [1.00 1.00 5.00 1.00]
// [1.00 1.00 5.00 1.00]
// ]

Tensor* t5 = Tensor::ones({3, 4});
Tensor* t6 = Tensor::full({3, 2}, 5.0f);
t5->set_select({"0:2", "-2:"}, t6);  // Set 3rd and 4th column from 1st and 2nd row to "5"
// [
// [1.00 1.00 5.00 5.00]
// [1.00 1.00 5.00 5.00]
// [1.00 1.00 1.00 1.00]
// ]

Expand

Tensor *Tensor::expand(int size)

Returns a new tensor with singleton dimensions expanded to a larger size.

Parameters: size – number of times to expand axis of dimension 1
Returns: void

Example:

 Tensor* t1 = new Tensor( {1, 2, 3}, {3, 1});

 Tensor* new_t = t1->expand(3);

// Other ways
Tensor::expand(t1, size); // static

Sorting

sort

Tensor *Tensor::sort(bool descending = false, bool stable = true)

Sort a tensor.

Parameters

descending – Wether to sort the tensor descending or not.
stable – Wether to use stable sorting or not. Stable sorting keeps the order of equal elements.

Returns

A tensor with the sorted elements.

Example:

 Tensor* t1 = Tensor::randn({5});
// [-0.01 0.34 0.10 -0.57 -0.28]

 Tensor* t2 = t1->sort();  // Ascending
// [-0.57 -0.28 -0.01 0.10 0.34]

 Tensor* t3 = t1->sort(true);  // Descending==True
// [0.34 0.10 -0.01 -0.28 -0.57]

// Other ways
t1->sort_();  // In-place
Tensor::sort(t1, t2); // static

argsort

Tensor *Tensor::argsort(bool descending = false, bool stable = true)

Sort the indices of a tensor according to the elements in each position.

Parameters

descending – Wether to sort the tensor descending or not.
stable – Wether to use stable sorting or not. Stable sorting keeps the order of equal elements.

Returns

A tensor with the sorted indices.

Example:

 Tensor* t1 = Tensor::randn({5});
// [-0.01 0.34 0.10 -0.57 -0.28]

 Tensor* t2 = t1->argsort();  // Ascending
// [3 4 0 2 1]

 Tensor* t3 = t1->argsort(true);  // Descending==True
// [1 2 0 4 3]

// Other ways
Tensor::argsort(t1, t2); // static