How to use functions like np.unique() and np.any() on pybind11::array?

[sun1638650145]

I want to use functions similar to np.unique() and np.any() on pybind11::array, so I first cast pybind11::array to Eigen::ArrayXf, Then proceed with processing. But no matter how I operate, I can only process 1/2-D array, but I also expect to be able to process 3D array, is it my idea is wrong, or pybind11 provides built-in functions.

void MyUnique(const pybind11::array &array) {
    auto arr = pybind11::cast<Eigen::ArrayXf>(array);

    // Then use std::unique and std::sort to remove duplicates.
}

Below is the function I expect.

unique_ = np.unique(y)
any_ = np.any(y != y.astype(int))

English is not my native language; please excuse typing errors.

[Skylion007]

You can always just import numpy through it's python interface to access it's advanced functionality.

auto np = py::module_::import("numpy");
auto out = np.attr("unique")(array);

See the documentation on importing Python modules for mode details: https://pybind11.readthedocs.io/en/stable/advanced/embedding.html?highlight=import#importing-modules

Alternatively, Eigen has it's own unique and sort functions etc so I would just use those on the array object.

[sun1638650145]

First, thank you very much, but it seems that I did not express what I mean clearly. I expect is to use pybind11::array to achieve similar functions, not directly from numpy.

[Skylion007]

We don't have those functions exposed via the C++ API through py:array.

[sun1638650145]

Because I just need to remove the duplicated values ​​of the pybind11::array, and don’t need the original shape, can we cast the array to a flat std::vector?

[Skylion007]

Yes, there is a caster from array to std::vector. It also has STL compatible iterators so you can call std::unique etc...

[sun1638650145]

I tried to convert to std::vector but 2-D and 3-D does not seem to be able to use pybind11::cast. What should I do?