check_sparse_tensor_invariants — PyTorch 2.0 documentation (original) (raw)

class torch.sparse.check_sparse_tensor_invariants(enable=True)[source]¶

A tool to control checking sparse tensor invariants.

The following options exists to manage sparsr tensor invariants checking in sparse tensor construction:

Using a context manager:
with torch.sparse.check_sparse_tensor_invariants():
run_my_model()
Using a procedural approach:
prev_checks_enabled = torch.sparse.check_sparse_tensor_invariants.is_enabled()
torch.sparse.check_sparse_tensor_invariants.enable()
run_my_model()
if not prev_checks_enabled:
torch.sparse.check_sparse_tensor_invariants.disable()
Using function decoration:
@torch.sparse.check_sparse_tensor_invariants()
def run_my_model():
...
run_my_model()
Using check_invariants keyword argument in sparse tensor constructor call. For example:

torch.sparse_csr_tensor([0, 1, 3], [0, 1], [1, 2], check_invariants=True)
Traceback (most recent call last):
File "", line 1, in
RuntimeError: crow_indices[..., -1] == nnz is not satisfied.

static disable()[source]¶

Disable sparse tensor invariants checking in sparse tensor constructors.

static enable()[source]¶

Enable sparse tensor invariants checking in sparse tensor constructors.

Note

By default, the sparse tensor invariants checks are disabled. Usetorch.sparse.check_sparse_tensor_invariants.is_enabled() to retrieve the current state of sparse tensor invariants checking.

Note

The sparse tensor invariants check flag is effective to all sparse tensor constructors, both in Python and ATen.

The flag can be locally overridden by the check_invariantsoptional argument of the sparse tensor constructor functions.

static is_enabled()[source]¶

Returns True if the sparse tensor invariants checking is enabled.

Note