Files
pytorch-lightning/pytorch_lightning/trainer/supporters.py
T
ddbf7de6dc Added accumulation of loggers' metrics for the same steps (#1278)
* `add_argparse_args` method fixed (argument types added)

* autopep8 fixes

* --gpus=0 removed from test (for ci tests)

* Update pytorch_lightning/trainer/trainer.py

Co-Authored-By: Joe Davison <joe@huggingface.co>

* test_with_accumulate_grad_batches added

* agg_and_log_metrics logic added to the base logger class

* small format fix

* agg metrics strategies removed (not to complicate stuff)

* agg metrics: handle zero step

* autopep8

* changelog upd

* flake fix

* metrics aggregators factored out, metrics_agg.py added + tests

* metrics agg default value added

* Update pytorch_lightning/loggers/metrics_agg.py

Co-Authored-By: Jirka Borovec <Borda@users.noreply.github.com>

* metrics aggregators factored out, metrics_agg.py added + tests

* metrics agg default value added

* Update pytorch_lightning/loggers/metrics_agg.py

Co-Authored-By: Jirka Borovec <Borda@users.noreply.github.com>

* remove .item which causes sync issues (#1254)

* remove .item which causes sync issues

* fixed gradient acc sched

* fixed gradient acc sched

* test_metrics_agg.py removed (all tested in doctrings), agg metrics refactored

* test_metrics_agg.py removed (all tested in doctrings), agg metrics refactored

* autopep8

* loggers base.py types fixed

* test

* test

* metrics aggregation for loggers: each key now has a specific function (or default one)

* metrics aggregation for loggers: each key now has a specific function (or default one)

* docstrings upd

* manual typehints removed from docstrings

* batch_size decreased for test `test_with_accumulate_grad_batches`

* extend running accum

* refactor

* fix tests

* fix tests

* allowed_types generator scoped

* trainer.py distutils was imported twice, fixed

* TensorRunningAccum refactored

* TensorRunningAccum added to change log (Changed)

* change log pull link added

Co-authored-by: Joe Davison <joe@huggingface.co>
Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>
Co-authored-by: William Falcon <waf2107@columbia.edu>
Co-authored-by: J. Borovec <jirka.borovec@seznam.cz>
2020-04-08 08:35:47 -04:00

76 lines
2.4 KiB
Python

import torch
class TensorRunningAccum(object):
"""Tracks a running accumulation values (min, max, mean) without graph
references.
Examples:
>>> accum = TensorRunningAccum(5)
>>> accum.last(), accum.mean()
(None, None)
>>> accum.append(torch.tensor(1.5))
>>> accum.last(), accum.mean()
(tensor(1.5000), tensor(1.5000))
>>> accum.append(torch.tensor(2.5))
>>> accum.last(), accum.mean()
(tensor(2.5000), tensor(2.))
>>> accum.reset()
>>> _= [accum.append(torch.tensor(i)) for i in range(13)]
>>> accum.last(), accum.mean(), accum.min(), accum.max()
(tensor(12.), tensor(10.), tensor(8.), tensor(12.))
"""
def __init__(self, window_length: int):
self.window_length = window_length
self.memory = torch.Tensor(self.window_length)
self.current_idx: int = 0
self.last_idx: int = None
self.rotated: bool = False
def reset(self) -> None:
"""Empty the accumulator."""
self = TensorRunningAccum(self.window_length)
def last(self):
"""Get the last added element."""
if self.last_idx is not None:
return self.memory[self.last_idx]
def append(self, x):
"""Add an element to the accumulator."""
# ensure same device and type
if self.memory.device != x.device or self.memory.type() != x.type():
x = x.to(self.memory)
# store without grads
with torch.no_grad():
self.memory[self.current_idx] = x
self.last_idx = self.current_idx
# increase index
self.current_idx += 1
# reset index when hit limit of tensor
self.current_idx = self.current_idx % self.window_length
if self.current_idx == 0:
self.rotated = True
def mean(self):
"""Get mean value from stored elements."""
return self._agg_memory('mean')
def max(self):
"""Get maximal value from stored elements."""
return self._agg_memory('max')
def min(self):
"""Get minimal value from stored elements."""
return self._agg_memory('min')
def _agg_memory(self, how: str):
if self.last_idx is not None:
if self.rotated:
return getattr(self.memory, how)()
else:
return getattr(self.memory[:self.current_idx], how)()