Function NdarraysToSparseCSFTensor

python/pyarrow/src/arrow/python/numpy_convert.cc:472–524

}

Status NdarraysToSparseCSFTensor(MemoryPool* pool, PyObject* data_ao, PyObject* indptr_ao,
                                 PyObject* indices_ao, const std::vector<int64_t>& shape,
                                 const std::vector<int64_t>& axis_order,
                                 const std::vector<std::string>& dim_names,
                                 std::shared_ptr<SparseCSFTensor>* out) {
  if (!PyArray_Check(data_ao)) {
    return Status::TypeError("Did not pass ndarray object for data");
  }
  const int ndim = static_cast<const int>(shape.size());
  PyArrayObject* ndarray_data = reinterpret_cast<PyArrayObject*>(data_ao);
  std::shared_ptr<Buffer> data = std::make_shared<NumPyBuffer>(data_ao);
  ARROW_ASSIGN_OR_RAISE(
      auto type_data,
      GetTensorType(reinterpret_cast<PyObject*>(PyArray_DESCR(ndarray_data))));

  std::vector<std::shared_ptr<Tensor>> indptr(ndim - 1);
  std::vector<std::shared_ptr<Tensor>> indices(ndim);

  for (int i = 0; i < ndim - 1; ++i) {
#ifdef Py_GIL_DISABLED
    PyObject* item = PySequence_ITEM(indptr_ao, i);
    RETURN_IF_PYERROR();
    OwnedRef item_ref(item);
#else
    PyObject* item = PySequence_Fast_GET_ITEM(indptr_ao, i);
#endif
    if (!PyArray_Check(item)) {
      return Status::TypeError("Did not pass ndarray object for indptr");
    }
    RETURN_NOT_OK(NdarrayToTensor(pool, item, {}, &indptr[i]));
    ARROW_CHECK_EQ(indptr[i]->type_id(), Type::INT64);  // Should be ensured by caller
  }

  for (int i = 0; i < ndim; ++i) {
#ifdef Py_GIL_DISABLED
    PyObject* item = PySequence_ITEM(indices_ao, i);
    RETURN_IF_PYERROR();
    OwnedRef item_ref(item);
#else
    PyObject* item = PySequence_Fast_GET_ITEM(indices_ao, i);
#endif
    if (!PyArray_Check(item)) {
      return Status::TypeError("Did not pass ndarray object for indices");
    }
    RETURN_NOT_OK(NdarrayToTensor(pool, item, {}, &indices[i]));
    ARROW_CHECK_EQ(indices[i]->type_id(), Type::INT64);  // Should be ensured by caller
  }

  auto sparse_index = std::make_shared<SparseCSFIndex>(indptr, indices, axis_order);
  *out = std::make_shared<SparseTensorImpl<SparseCSFIndex>>(sparse_index, type_data, data,
                                                            shape, dim_names);
  return Status::OK();
}

Status NdarraysToSparseCSRMatrix(MemoryPool* pool, PyObject* data_ao, PyObject* indptr_ao,
                                 PyObject* indices_ao, const std::vector<int64_t>& shape,