row-major or col-major?

[rogersce]

row-major or col-major?

[RunOrVeith]

For anyone else stumbling across this:
Numpy stores data in row-major by default, but Eigen in column-major.

So in order to load the data correctly, you have to do the following

    cnpy::npz_t loadedNpz = cnpy::npz_load(file.string());
    cnpy::NpyArray npPositions = loadedNpz["positions"];

    // YOU HAVE TO BE CAREFUL, NUMPY STORES DATA IN ROW-MAJOR, BUT EIGEN in COLUMN-MAJOR BY DEFAULT
    float *positionsArray = npPositions.data<float>();
    Eigen::MatrixX3f positions;
    positions = Eigen::Map<Eigen::Matrix<float, Eigen::Dynamic, 3, Eigen::RowMajor>>(positionsArray, npPositions.shape[0], npPositions.shape[1]);

If you load something that is square (e.g. a 4x4 matrix), it will appear to be transposed if you don't specify the Eigen::RowMajor flag. If you load something non-square (e.g 100x3), it will be completely different to the original data order if you don't specify the flag.

[firdota]

if the var "fortran_order" in struct NpyArray is TRUE, then col-oder; else row-oder

[poor1017]

Thank you two.

[csyhping]

For anyone else stumbling across this: Numpy stores data in row-major by default, but Eigen in column-major.

So in order to load the data correctly, you have to do the following

    cnpy::npz_t loadedNpz = cnpy::npz_load(file.string());
    cnpy::NpyArray npPositions = loadedNpz["positions"];

    // YOU HAVE TO BE CAREFUL, NUMPY STORES DATA IN ROW-MAJOR, BUT EIGEN in COLUMN-MAJOR BY DEFAULT
    float *positionsArray = npPositions.data<float>();
    Eigen::MatrixX3f positions;
    positions = Eigen::Map<Eigen::Matrix<float, Eigen::Dynamic, 3, Eigen::RowMajor>>(positionsArray, npPositions.shape[0], npPositions.shape[1]);

If you load something that is square (e.g. a 4x4 matrix), it will appear to be transposed if you don't specify the Eigen::RowMajor flag. If you load something non-square (e.g 100x3), it will be completely different to the original data order if you don't specify the flag.

hi @RunOrVeith , i tried with both Eigne::RowMajor and Eigen::ColMajor, the results are same. I need to call .transpose() to get the same data as numpy. Could you help with this? Thanks.