ancri

It's pleasant when code lines up. Recording here oft-used-together words have the same length:

train
valid

prediction from
actual value in

pred
resp

ancri

(optional) set up pubkey authorization 
(taken from https://unix.stackexchange.com/questions/23291/how-to-ssh-to-remote-server-using-a-private-key)

$ scp -p your_pub_key.pub user@host:
$ ssh user@host
$ cat id_dsa.pub >> ~/.ssh/authorized_keys

make sure this is in ~/.ssh/config on the remote:
PubkeyAuthentication yes

ancri

from contextlib import contextmanager

@contextmanager
def optional(condition, context_manager):
    # taken from https://stackoverflow.com/a/41251962/12558195
    if condition:
        with context_manager:
            yield
    else:
        yield

ancri

# Below is a cheeky little utility that helps you 
# avoid doing all the tedious math by piggy-backing 
# on torch’s error messages. The assumption is that 
# net is your Module which is composed of a sequence 
# of input shape-agnostic layers (such as Conv & ReLU 
# & BatchNorm & MaxPool layers), followed at some 
# point by a flattening and a linear layer whose required 
# size you’re trying to figure out. height and width 
# refer to your desired input image shape.