Logger

%%capture
!pip install labml

from labml import logger
from labml.logger import Text, Color

You can log to the screen with log().

Logging with colors

logger.log("Colors are missing when viewed on github", Text.highlight)

Colors are missing when viewed on github

You can use predifined styles

logger.log([
    ('Styles ', Text.heading),
    ('Danger ', Text.danger),
    ('Warning ', Text.warning),
    ('Meta ', Text.meta),
    ('Key ', Text.key),
    ('Meta2 ', Text.meta2),
    ('Title ', Text.title),
    ('Heading ', Text.heading),
    ('Value ', Text.value),
    ('Highlight ', Text.highlight),
    ('Subtle', Text.subtle)
])

Styles Danger Warning Meta Key Meta2 Title Heading Value Highlight Subtle

Or, specify colors

logger.log([
    ('Colors ', Text.heading),
    ('Red ', Color.red),
    ('Black ', Color.black),
    ('Blue ', Color.blue),
    ('Cyan ', Color.cyan),
    ('Green ', Color.green),
    ('Orange ', Color.orange),
    ('Purple Heading ', [Color.purple, Text.heading]),
    ('White', Color.white),
])

Colors Red Black Blue Cyan Green Orange Purple Heading White

Logging debug info

You can pretty print python objects with inspect().

logger.inspect(a=2, b=1)

a: 2
b: 1

logger.inspect(dict(name='Name', price=22))

 name: "Name"
price: 22
Total 2 item(s)

Log PyTorch tensors and NumPy arrays

import torch

torch_tensor = torch.arange(0, 100).view(10, 10)
logger.inspect(torch_tensor)

dtype: torch.int64
shape: [10, 10]
min: 0 max: 99 mean:  49.5000 std:  29.0115
[
 [0, 1, 2, ..., 9 ], 
 [10, 11, 12, ..., 19 ], 
 [20, 21, 22, ..., 29 ], 
 ..., 
 [90, 91, 92, ..., 99 ]
]

numpy_array = torch_tensor.numpy()
logger.inspect(numpy_array)

dtype: int64
shape: [10, 10]
min: 0 max: 99 mean:  49.5000 std:  28.8661
[
 [0, 1, 2, ..., 9 ], 
 [10, 11, 12, ..., 19 ], 
 [20, 21, 22, ..., 29 ], 
 ..., 
 [90, 91, 92, ..., 99 ]
]