Generalize task_t for generic 3D tensors

accosmin / nano

C++ library [machine learning & numerical optimization] - superseeded by libnano

MIT License

1 stars 0 forks source link

Generalize task_t for generic 3D tensors #96

Closed accosmin closed 8 years ago

accosmin commented 8 years ago

Currently the library can process only images, but it can be extended to process generic 3D tensors.

accosmin commented 8 years ago

Should also take some time to extend the tensor module to have explicit 3D and 4D tensors. The Tensor module from Eigen3 is still unsupported...

accosmin commented 8 years ago

Also create supporting synthetic tasks: e.g. learn mapping of the form {f(x) + noise} where the mapping type (linear, quadratic ...) and the noise ratio can be parameters of the task.

accosmin commented 8 years ago

To implement:

tensor3D: dims(), rows(), cols(), data(), vector(), vector(i), matrix(i), planeSize()
tensor4D: dim1(), dim2(), rows(), cols(), data(), vector(), vector(i, j), matrix(i, j), planeSize()
map_tensor3d, map_tensor4d from raw pointers...