feth / monologue Goto Github PK

At first, we need this poc to work :

from logging import Logger, StreamHandler, Formatter
import sys

logger = Logger("name")

handler = StreamHandler(sys.stdout)

class MyFormatter(Formatter):
    pass


myformatter = MyFormatter()
handler.setFormatter(myformatter)

logger.addHandler(handler)



class Bidon(object):
    """
    >>> b = Bidon()
    [Bidon] coucou
    """
    def __init__(self):
        logger.warning("coucou")

While we're talking progress information, I was running a long computation by a sloppy friend coder of mine the other day, and there was no way to tell when I'd be able to get my machine back^W^W^W^Whand the result.

ETA stands for "estimated time of arrival". It belongs to the same family of features as percent completion and might look approx. like this:

[11/02/2012 18h14] [Long calculus] starting 100000 iterations
[11/02/2012 18h32] [Long calculus] 10% complete - 10000 iterations done - Estimated completion in 02h42m23s at 21h14
[11/02/2012 18h53] [Long calculus] 20% complete - 20000 iterations done - Estimated completion in 02h25m00s at 21h18
...

Some attributes of the main object are managed by accessors.
Making these attributes public and removing accessors would reduce the cross section of this main object.

We're looking for API simplification whenever possible.

Main features:

• displays progress
• displays log messages

So the class was named ProgressAndLog and the main method get_logger.
Do you have a better idea? What do you think of the boring «reporter»?

Integrate docstrings into documentation
title says it.

Title says it all

usage scope: scikit functions.

the 'check_logger' function would accept all current accepted values for verbose PLUS a logger.

returns: a logger

maybe provide a start() that accepts an optional id and returns it.

complete would provide a separator, the time elapsed, the iterations number (if relevant)

for instance

...........................................................................

or

=======================================

@GaelVaroquaux

As we inherit logging.Logger, there are about 30+ public methods in the main class.
I'm not sure it is too many, what do you think?

that's a feeling @GaelVaroquaux has.

Do we prefer an API with

logger = get_logger()

or

log("coucou")

or both ?

For the latter, we could use a default_logger() method, that would enable the log method.

When detecting ipython in Qt mode : provide progressbars instead of dots (and perhaps dots)

get_logger should not require a name, but optionnaly accept a prefix.

When not specified, the prefix should be calculated. See #10

I get the above AttributeError when running the following example (from doc/manual.rst) :

# Configure: a dot every 10 steps
logger.dot_every(10)

# Configure: progress message every 100 steps
logger.progress_every(100)

# Optional: reset the number of iterations
logger.progress_reset()

while x != 1:

    # count one step
    logger.step()

    if x % 2 == 0:  # if x is odd
        x /= 2  # halve x
    else:
        x = 3 * x + 1

logger.complete()

Copy and customize the target (joblib) documentation infrastructure (Makefile etc).