def plot(self):
if self.ax is None:
self.ax = pl.figure().add_subplot(111)
ax = self.ax
with update_ax(ax):
sty = self.sty
for k, v in self.baselines.items():
ax.axhline(v, label=k, **sty[k])
data = self.data
for k, v in data.iteritems():
xs, ys = zip(*data[k])
if self.averaging:
sty[k]['alpha'] = 0.5
[l] = ax.plot(xs, ys, label=k, **sty[k])
# show a dot in addition to the line
if 0:
s = sty[k].copy()
s['lw'] = 0
s['c'] = l.get_color()
ax.scatter(xs, ys, label=k, **s)
if self.averaging:
# TODO: averaging doesn't support irregular time steps.
halflife = 10
_,yy = zip(*data[k])
ax.plot(xs,
pandas.Series(yy).ewm(halflife=halflife).mean(),
lw=2, c=l.get_color())
ax.set_title(self.name)
if self.legend:
ax.legend(loc=4)
def draw(self):
ax = self.ax
with update_ax(ax):
sty = self.sty
for k, v in self.baselines.items():
ax.axhline(v, label=k, **sty[k])
data = self.data
for k, v in data.items():
xs, ys = np.array(data[k]).T
if self.smoothing is not None:
# mute the raw signal when we are smoothing.
sty[k]['alpha'] = 0.5
[l] = ax.plot(xs, ys, label=k, **sty[k])
c = l.get_color()
# show a dot in addition to the line
if 0:
s = sty[k].copy()
s['lw'] = 0
s['c'] = c
ax.scatter(xs, ys, label=k, **s)
self.draw_smoothing(xs, ys, c=c)
self.draw_bands(xs, ys, c=c)
if self.xscale: ax.set_xscale(self.xscale)
if self.yscale: ax.set_yscale(self.yscale)
if self.name: ax.set_title(self.name)
if self.legend: ax.legend(loc='best')
self.draw_extra(ax)
return self
def plot(self):
if self.ax is None:
self.ax = pl.figure().add_subplot(111)
ax = self.ax
with update_ax(ax):
sty = self.sty
for k, v in self.baselines.items():
ax.axhline(v, label=k, **sty[k])
data = self.data
for k, v in data.iteritems():
xs, ys = zip(*data[k])
if self.averaging:
sty[k]['alpha'] = 0.5
[l] = ax.plot(xs, ys, label=k, **sty[k])
# show a dot in addition to the line
if 0:
s = sty[k].copy()
s['lw'] = 0
s['c'] = l.get_color()
ax.scatter(xs, ys, label=k, **s)
if self.averaging:
# TODO: averaging doesn't support irregular time steps.
halflife = 10
_, yy = zip(*data[k])
ax.plot(xs,
pandas.Series(yy).ewm(halflife=halflife).mean(),
lw=2,
c=l.get_color())
ax.set_title(self.name)
if self.legend:
ax.legend(loc=4)
def plot(self):
if self.ax is None:
self.ax = pl.figure().add_subplot(111)
ax = self.ax
with update_ax(ax):
sty = self.sty
for k, v in self.baselines.items():
ax.axhline(v, label=k, **sty[k])
data = self.data
for k, v in data.iteritems():
xs, ys = np.array(data[k]).T
if self.averaging:
sty[k]['alpha'] = 0.5
[l] = ax.plot(xs, ys, label=k, **sty[k])
# show a dot in addition to the line
if 0:
s = sty[k].copy()
s['lw'] = 0
s['c'] = l.get_color()
ax.scatter(xs, ys, label=k, **s)
if self.averaging:
s = pandas.Series(ys)
if 0:
halflife = 20
r = s.ewm(halflife=halflife)
M = r.mean()
s = r.std()
U = M + 2 * s
L = M - 2 * s
else:
window = min(len(ys), 10)
r = s.rolling(window, min_periods=0)
M = r.median()
#U = r.max()
#L = r.min()
U = r.quantile(.9)
L = r.quantile(.1)
ax.plot(xs, M, lw=2, c=l.get_color())
ax.fill_between(xs, U, L, color=l.get_color(), alpha=0.25)
if self.xscale: ax.set_xscale(self.xscale)
if self.yscale: ax.set_yscale(self.yscale)
if self.name: ax.set_title(self.name)
if self.legend: ax.legend(loc='best')
def draw(self):
if self.ax is None:
self.ax = pl.figure().add_subplot(111)
ax = self.ax
with update_ax(ax):
sty = self.sty
for k, v in self.baselines.items():
ax.axhline(v, label=k, **sty[k])
data = self.data
for k, v in data.items():
xs, ys = np.array(data[k]).T
if self.smoothing is not None:
sty[k]['alpha'] = 0.5
[l] = ax.plot(xs, ys, label=k, **sty[k])
# show a dot in addition to the line
if 0:
s = sty[k].copy()
s['lw'] = 0
s['c'] = l.get_color()
ax.scatter(xs, ys, label=k, **s)
if self.smoothing is not None:
s = pandas.Series(ys)
if 0:
halflife = 20
r = s.ewm(halflife=halflife)
M = r.mean()
s = r.std()
U = M + 2*s
L = M - 2*s
else:
window = min(len(ys), self.smoothing)
r = s.rolling(window, min_periods=0)
M = r.median()
#U = r.max()
#L = r.min()
U = r.quantile(.9)
L = r.quantile(.1)
ax.plot(xs, M, lw=2, c=l.get_color())
ax.fill_between(xs, U, L, color=l.get_color(), alpha=0.25)
if self.xscale: ax.set_xscale(self.xscale)
if self.yscale: ax.set_yscale(self.yscale)
if self.name: ax.set_title(self.name)
if self.legend: ax.legend(loc='best')
def hinton(ax, W, gold, got, maxWeight=None, title=''):
"""Visualize pruning mask and parser output. Produces a image which is similar
to an (upper-triangular) Hinton diagram.
Size and color indicate pruning strength and whether hard decisions is
keep/prune (white/black, respectively).
Additional annotations
* gold box -- unlabled span in "gold" standard tree
* Parser output
- green border: unlabeled item is correct.
- red border: unlabeled is not correct.
TODO:
- How do I indicate errors in current parse? current visualization only
shows which unlabeled spans are good? In many cases unlabeled recall is
perfect, but labeled recall is quite low, e.g., 60%.
"""
N = W.shape[0]
if ax is None:
ax = pl.figure().add_subplot(111)
def blob(x, y, area, c):
"""
Draws a square-shaped blob with the given area (< 1) at
the given coordinates.
"""
hs = sqrt(area) / 2
xcorners = array([x - hs, x + hs, x + hs, x - hs])
ycorners = array([y - hs, y - hs, y + hs, y + hs])
ax.fill(xcorners, ycorners, c, edgecolor=c)
with update_ax(ax):
height, width = W.shape
if maxWeight is None:
maxWeight = np.max(np.abs(W)) / .5
ax.fill(array([1, width, width, 1]),
array([0, 0, height, height]), 'gray')
ax.axis('off')
ax.axis('equal')
for y in xrange(height):
for x in xrange(y, width):
_x = x+1
_y = y+1
w = W[y,x]
if abs(x-y) > 1 and abs(x-y) != N:
if (y, x) in got:
if (y,x) in gold:
c = 'green'
else:
c = 'red'
blob(_x - 0.5, height - _y + 0.5, 0.85, c)
if (y, x) in gold or abs(x-y) == 1 or abs(x-y) == N:
blob(_x - 0.5, height - _y + 0.5, 0.70, 'yellow')
if abs(x - y) in (0, 1, height):
continue
if w >= 0:
blob(_x - 0.5, height - _y + 0.5, min(1, w / maxWeight), 'white')
else:
blob(_x - 0.5, height - _y + 0.5, min(1, -w / maxWeight), 'black')
if title:
ax.set_title(title)