我有一个简单的问题:如何将内置的Python日志记录器的print
函数更改为tqdm.write
,以便日志记录消息不会干扰tqdm的进度条?
tqdm
现在具有内置的上下文管理器,用于重定向记录器:
import logging
from tqdm import trange
from tqdm.contrib.logging import logging_redirect_tqdm
LOG = logging.getLogger(__name__)
if __name__ == '__main__':
logging.basicConfig(level=logging.INFO)
with logging_redirect_tqdm():
for i in trange(9):
if i == 4:
LOG.info("console logging redirected to `tqdm.write()`")
# logging restored
从tqdm文档复制
LOG
设置了自己的格式化程序,则应该执行 logging_redirect_tqdm(loggers=[LOG])
。这将导致 tqdm
使用 LOG
的格式化程序。 - Daniel Walkertqdm.write
,都需要一个上下文管理器,而另一种解决方案可以用于全局更改。 - Guillochonimport logging
import tqdm
class TqdmLoggingHandler(logging.Handler):
def __init__(self, level=logging.NOTSET):
super().__init__(level)
def emit(self, record):
try:
msg = self.format(record)
tqdm.tqdm.write(msg)
self.flush()
except Exception:
self.handleError(record)
然后将这个添加到日志链中:
import time
log = logging.getLogger(__name__)
log.setLevel(logging.INFO)
log.addHandler(TqdmLoggingHandler())
for i in tqdm.tqdm(range(100)):
if i == 50:
log.info("Half-way there!")
time.sleep(0.1)
编辑:在评论中勤奋的读者@BlaineRogers指出了一个调用super TqdmLoggingHandler的init方法的错误,已经进行了修复。(如果有人想进一步了解Python这个模糊领域,请参考https://fuhm.net/super-harmful/)
import tqdm
而不是from tqdm import tqdm
,否则IO会中断进度条。 - Alexander McFarlaneTqdmLoggingHandler
的子类,则super(self.__class__, self).__init__(level)
会导致RecursionError
。super(cls, self).__init__()
调用self.__class__
方法解析顺序中cls
之后的类的__init__
方法。假设MyTqdmLoggingHandler
是TqdmLoggingHandler
的子类,则MyTqdmLoggingHandler()
调用TqdmLoggingHandler.__init__(self)
,该方法调用super(MyTqdmLoggingHandler, self).__init__()
,进而调用TqdmLoggingHandler.__init__(self)
。 - Blaine Rogerstqdm.notebook.tqdm.write
而不是tqdm.tqdm.write
。 - robertspierre基于RolKau上面的回答,简化版:
import logging
from tqdm import tqdm
class TqdmLoggingHandler(logging.StreamHandler):
"""Avoid tqdm progress bar interruption by logger's output to console"""
# see logging.StreamHandler.eval method:
# https://github.com/python/cpython/blob/d2e2534751fd675c4d5d3adc208bf4fc984da7bf/Lib/logging/__init__.py#L1082-L1091
# and tqdm.write method:
# https://github.com/tqdm/tqdm/blob/f86104a1f30c38e6f80bfd8fb16d5fcde1e7749f/tqdm/std.py#L614-L620
def emit(self, record):
try:
msg = self.format(record)
tqdm.write(msg, end=self.terminator)
except RecursionError:
raise
except Exception:
self.handleError(record)
测试:
import time
log = logging.getLogger(__name__)
log.setLevel(logging.INFO)
log.addHandler(TqdmLoggingHandler())
# ^-- Assumes this will be the unique handler emitting messages to sys.stdout.
# If other handlers output to sys.stdout (without tqdm.write),
# progress bar will be interrupted by those outputs
for i in tqdm(range(20)):
log.info(f"Looping {i}")
time.sleep(0.1)
<注意:如果您正在使用Jupyter笔记本工作,进度条将会被中断,并且据我所知没有避免的方法。>
一种简单但不太优雅的解决方案是将tqdm对象转换为字符串。之后,您可以记录消息(或按照您的需求处理消息)。"format_dict"属性也可能有用:
from tqdm import tqdm
import time
#loop with progressbar:
it=tqdm(total=10)
for i in range(10):
time.sleep(0.1)
it.update(1)
it.close()
print("\n","--"*10)
# Convert tdqm object last output to sting
str_bar_msg = str(it)
print(str_bar_msg)
# See attributes:
print(it.format_dict)
输出:
100%|██████████| 10/10 [00:01<00:00, 8.99it/s]
--------------------
100%|██████████| 10/10 [00:01<00:00, 8.98it/s]
{'n': 10, 'total': 10, 'elapsed': 1.1145293712615967, 'ncols': None, 'nrows': None, 'prefix': '', 'ascii': False, 'unit': 'it', 'unit_scale': False, 'rate': None, 'bar_format': None, 'postfix': None, 'unit_divisor': 1000, 'initial': 0, 'colour': None}
最好的问候
from tqdm.notebook import tqdm
。 - robertspierrefrom tqdm.auto import tqdm
来自动检测是否在Jupyter笔记本中,并使用适当的进度条。 - tbrugere最简单的方法是更改StreamHandler
对象的流,例如:
import logging
from tqdm import tqdm, trange
import time
log = logging.getLogger(__name__)
log.setLevel(logging.INFO)
handler = logging.StreamHandler()
handler.setStream(tqdm) # <-- important
handler = log.addHandler(handler)
for i in trange(100):
if i == 50:
log.info("Half-way there!")
time.sleep(0.1)
log.info
会出现警告,无法正常工作。 - krenerdlog.info("Half-way there!")
替换为 tqdm.write("Half-way there!")
。结果是相同的,这意味着问题出在 tqdm 模块本身,因为使用 .write(...)
是官方的打印方法。 - asiloisad新的io处理程序非常有用!
class TqdmToLogger(io.StringIO):
logger = None
level = None
buf = ""
def __init__(self, logger, level=None, mininterval=5):
super(TqdmToLogger, self).__init__()
self.logger = logger
self.level = level or logging.INFO
self.mininterval = mininterval
self.last_time = 0
def write(self, buf):
self.buf = buf.strip("\r\n\t ")
def flush(self):
if len(self.buf) > 0 and time.time() - self.last_time > self.mininterval:
self.logger.log(self.level, self.buf)
self.last_time = time.time()```
# before this line, you need to create logger with file handler
tqdm_out = TqdmToLogger(logger)
tbar = tqdm(sample, total=len(sample), file=tqdm_out)
logger.info("Model Inference.")
for it, batch_data in enumerate(tbar):
pass
```