TBWrapper

TBW is a wrapper to facilitate tensorboard logging. Below are some scripts with examples for different data types.

Note

Other data types (e.g., figure, histogram) are available.

SCALAR

import torch
from tbw import TBWrapper

wrapper = TBWrapper('logs') # path to logging directory
loss_writer = wrapper('scalar', 'loss')

for _ in range(10):
    loss_writer(torch.randn(1))

del wrapper

---

SCALARS

import torch
from tbw import TBWrapper

wrapper = TBWrapper('logs')
loss_writer = wrapper('scalars', 'losses')

for _ in range(10):
    losses_dict = {'loss_1': torch.randn(1),
                   'loss_2': torch.randn(1)}
    loss_writer(losses_dict)

del wrapper

---

IMAGE

import torch
from tbw import TBWrapper

wrapper = TBWrapper('logs')
loss_writer = wrapper('image', 'results')

for _ in range(10):
    loss_writer(torch.randn((3,128, 128)))

del wrapper

---

VIDEO

import torch
from tbw import TBWrapper

wrapper = TBWrapper('logs')
loss_writer = wrapper('video', 'results')

for _ in range(10):
    loss_writer(torch.randn((1, 4, 3,128, 128)), fps=2.0)

del wrapper

---

TEXT

import torch
from tbw import TBWrapper

wrapper = TBWrapper('logs')
text_writer = wrapper('text', 'results')

for i in range(10):
    loss_writer(f'text string {i}')

del wrapper

---

EMBEDDING

import torch
from tbw import TBWrapper

wrapper = TBWrapper('logs')
embedding_writer = wrapper('embedding', 'results')
embedding_writer(torch.randn(10,256), metadata=torch.arange(10))

del wrapper

---

MULTIPLE WRITERS

import torch
from tbw import TBWrapper, TBType

wrapper = TBWrapper('logs') # path to logging directory

# can directly use enum data types instead
wrapper(TBType.SCALAR, 'loss')
wrapper(TBType.IMAGE, 'image')
wrapper(TBType.VIDEO, 'video')

for _ in range(10):

    # Find writer objects by tag
    wrapper['loss'](torch.randn(1))
    wrapper['image'](torch.randn(3, 128, 128))

wrapper['video'](torch.randn(1, 5, 3, 128, 128))

#optional cleanup
wrapper['loss'].flush()
wrapper['image'].flush()
wrapper['video'].flush()
del wrapper

---