Nested Tensors Are Cool

disposable cameras are cool, but i dont really understand the point

---

source

unflatten_dim_from_batch

def unflatten_dim_from_batch(
    embs, c_in
):

---

source

flatten_dim_to_batch

def flatten_dim_to_batch(
    x, dim:int=1
):

---

source

return_nested_dim

def return_nested_dim(
    x
):

---

source

coerce_offsets

def coerce_offsets(
    src, tgt
):

See https://github.com/pytorch/pytorch/issues/138180 Coerce offsets to be the same for two nested tensors so element wise operations can occur.

use: src = coerce_offsets(src, tgt)

---

source

get_predictions_nested_survival

def get_predictions_nested_survival(
    data_loader, model, dataloader_name:str='', autocast:bool=True
):

---

source

get_predictions_nested

def get_predictions_nested(
    data_loader, model, dataloader_name:str='', autocast:bool=True
):

---

source

get_embeddings_nested

def get_embeddings_nested(
    data_loader, model, dataloader_name:str='', autocast:bool=True
):