import numpy as np
import pylab as pl
import pandas
from collections import defaultdict
from arsenal.viz.util import update_ax
from arsenal.misc import ddict
lc = ddict(lambda name: LearningCurve(name))
class LearningCurve(object):
"""
Plot learning curve as data arrives.
"""
def __init__(self, name, sty=None, legend=True, averaging=True):
self.name = name
self.baselines = {}
self.baselines_reward = {}
self.data = defaultdict(list)
self.sty = defaultdict(dict)
if sty is not None:
self.sty.update(sty)
self.ax = None
self.legend = legend
self.averaging = averaging
self.yscale = None
self.xscale = None
def plot(self):
if self.ax is None:
expect='expect',
got='got',
show_regression=1,
scatter=1,
**kw):
from arsenal.maths import compare
if self.ax is None:
self.ax = pl.figure().add_subplot(111)
if self.df.empty:
return
with update_ax(self.ax):
compare(expect, got, data=self.df).plot(ax=self.ax, **kw)
# Global references to numerical debugger class.
DEBUG = ddict(NumericalDebug)
@contextmanager
def lineplot(name,
with_ax=False,
halflife=20,
xlabel=None,
ylabel=None,
title=None,
**style):
with axman(name, xlabel=xlabel, ylabel=ylabel, title=title) as ax:
data = DATA[name]
if with_ax:
yield (data, ax)
else:
self.draw()
self.last_update = time()
return self
def should_update(self):
"Returns true if its been long enough (>= `min_time`) since the `last_update`."
return time() - self.last_update >= self.min_time
def __reduce__(self):
# Default pickle fails because of the reference to the plotting axis
x = self.__dict__.copy()
x['ax'] = None
return (LearningCurve, (self.name, self.sty), x)
lc = ddict(LearningCurve)
#def run():
#
# lc = LearningCurve('test')
# lc.smooth('ewm', 'mean', half_life = 0.01).bands('std')
# for t in range(1, 1000):
# lc.update(t, signal = np.exp(np.log(t) * -0.5 + np.random.randn()))
# if t % 10 == 0:
# lc.update(t, signal2 = np.exp(np.log(t) * -0.25 + np.random.randn()))
# lc.loglog().draw()
#
# pl.ioff(); pl.show()
#
#
#if __name__ == '__main__':
import numpy as np
import pylab as pl
import pandas
from collections import defaultdict
from arsenal.viz.util import update_ax
from arsenal.misc import ddict
lc = ddict(lambda name: LearningCurve(name))
from time import time
class LearningCurve(object):
"""
Plot learning curve as data arrives.
"""
def __init__(self, name, sty=None, legend=True, smoothing=10):
self.name = name
self.baselines = {}
self.data = defaultdict(list)
self.sty = defaultdict(dict)
if sty is not None:
self.sty.update(sty)
self.ax = None
self.legend = legend
self.yscale = None
self.xscale = None
self.last_update = time()
self.min_time = 0.5
def __init__(self, title):
self.title = title
self.timers = ddict(Timer)
def __init__(self, title):
self.title = title
self.timers = ddict(Timer)
"Pass in column values for the row by name as keyword arguments."
self._data.append(kw)
self.uptodate = False
return self
def compare(self, expect='expect', got='got', show_regression=1, scatter=1, **kw):
from arsenal.math import compare
if self.ax is None:
self.ax = pl.figure().add_subplot(111)
if self.df.empty:
return
with update_ax(self.ax):
compare(expect, got, data=self.df).plot(ax=self.ax, **kw)
# Global references to numerical debugger class.
DEBUG = ddict(NumericalDebug)
@contextmanager
def lineplot(name, with_ax=False, halflife=20, xlabel=None, ylabel=None, title=None, **style):
with axman(name, xlabel=xlabel, ylabel=ylabel, title=title) as ax:
data = DATA[name]
if with_ax:
yield (data, ax)
else:
yield data
ax.plot(range(len(data)), data, alpha=0.5, **style)
if halflife:
ax.plot(pd.Series(data).ewm(halflife=halflife).mean(), alpha=0.5, c='k', lw=2)
@contextmanager
def timers():
return ddict(Timer)
