def graph_rocchio(fname):
with open(fname) as f:
results = pickle.load(f)
for percent in results.keys():
x = []
y = []
z = []
for alpha in results[percent]:
for beta in results[percent][alpha]:
x.append(alpha)
y.append(beta)
recalls = results[percent][alpha][beta]['recalls']
z.append(sum(recalls) / len(recalls))
fig = plt.figure()
ax = fig.gca(projection='3d')
xi = np.linspace(min(x), max(x))
yi = np.linspace(min(y), max(y))
X, Y = np.meshgrid(xi, yi)
Z = griddata(x, y, z, xi, yi)
surf = ax.plot_surface(X, Y, Z, rstride=5, cstride=5, cmap=cm.jet,
linewidth=1, antialiased=True)
ax.set_zlim3d(np.min(Z), np.max(Z))
fig.colorbar(surf)
ax.set_xlabel('alpha')
ax.set_ylabel('beta')
ax.set_zlabel('recall')
plt.show()
def submit_time_histogram(arr):
"""
Use Matplotlib to plot a normalized histogram of submit times
"""
from math import ceil, log
try:
import matplotlib.mlab as mlab
from prettyplotlib import plt
except ImportError:
print(
'You must have Matplotlib and Prettyplotlib installed to plot a histogram.'
)
# Use Sturges' formula for number of bins: k = ceiling(log2 n + 1)
k = ceil(log(len(arr), 2) + 1)
n, bins, patches = plt.hist(arr,
k,
normed=1,
facecolor='green',
alpha=0.75)
# throw a PDF plot on top of it
#y = mlab.normpdf(bins, np.mean(arr), np.std(arr))
#l = plt.plot(bins, y, 'r--', linewidth=1)
# Get a Bayesian confidence interval for mean, variance, standard deviation
dmean, dvar, dsd = bayes_mvs(deltas)
# drop a line in at the mean for fun
plt.axvline(dmean[0], color='blue', alpha=0.5)
plt.axvspan(dmean[1][0], dmean[1][1], color='blue', alpha=0.5)
plt.axvline(np.median(deltas), color='y', alpha=0.5)
# Caclulate a Kernel Density Estimate
density = gaussian_kde(deltas)
xs = np.arange(0., np.max(deltas), 0.1)
density.covariance_factor = lambda: .25
density._compute_covariance()
plt.plot(xs, density(xs), color='m')
#FIXME: come up with better legend names
#plt.legend(('Normal Curve', 'Mean', 'Median', 'KDE'))
plt.legend(('Mean', 'Median', 'KDE'))
plt.xlabel('Submit Times (in Seconds)')
plt.ylabel('Probability')
plt.title('Histogram of Worker submit times')
plt.grid(True)
plt.show()
def draw_color_mesh(self):
'''
Draws a heatmap of pairs of heroes which co-occur
in the winning and in the losing teams, useful to
visualize the relationship between strong pairs of
heroes which lead to victories vs. weak pairs of
heroes which don't have much synergy
'''
red_yellow = brewer2mpl.get_map('YlGnBu', 'Sequential', 9).mpl_colormap
fig, ax = plt.subplots(1, figsize=(13, 10))
ax.set_xlim([0, self.c])
ax.set_ylim([0, self.c])
mesh = np.zeros((self.c, self.c), dtype=float)
for i in range(0, self.c):
for j in range(0, self.c):
if i >= j:
# Same hero cannot be picked twice
continue
if (i, j) in self.dwstat:
if self.ddstat[(i, j)] != 0:
k = round(
self.dwstat[(i, j)] /
float(self.ddstat[(i, j)] + self.dwstat[(i, j)]),
2)
mesh[i][j] = k
mesh[j][i] = k
# *************************************************************** #
# Code to calculate the max ratios in the heatmap
# and obtain their hero indices too
# Get the indices for the largest `num_largest` values.
num_largest = 8
indices = mesh.argpartition(mesh.size - num_largest,
axis=None)[-num_largest:]
x, y = np.unravel_index(indices, mesh.shape)
print "full:"
print "x =", x
print "y =", y
print "Largest values:", mesh[x, y]
# print "Compare to: ", np.sort(mesh, axis=None)[-num_largest:]
# **************************************************************** #
ppl.pcolormesh(fig, ax, mesh, cmap=red_yellow)
fig.savefig('../Figures/HeatMap-heroPairs.png')
plt.show()
plt.clf()
def show_timeorder_info(Dt, mesh_sizes, errors):
'''Performs consistency check for the given problem/method combination and
show some information about it. Useful for debugging.
'''
# Compute the numerical order of convergence.
orders = {}
for key in errors:
orders[key] = _compute_numerical_order_of_convergence(Dt, errors[key])
# Print the data to the screen
for i, mesh_size in enumerate(mesh_sizes):
print
print('Mesh size %d:' % mesh_size)
print('dt = %e' % Dt[0]),
for label, e in errors.items():
print(' err_%s = %e' % (label, e[i][0])),
print
for j in range(len(Dt) - 1):
print(' '),
for label, o in orders.items():
print(' ord_%s = %e' % (label, o[i][j])),
print
print('dt = %e' % Dt[j+1]),
for label, e in errors.items():
print(' err_%s = %e' % (label, e[i][j+1])),
print
# Create a figure
for label, err in errors.items():
pp.figure()
ax = pp.axes()
# Plot the actual data.
for i, mesh_size in enumerate(mesh_sizes):
pp.loglog(Dt, err[i], '-o', label=mesh_size)
# Compare with order curves.
pp.autoscale(False)
e0 = err[-1][0]
for o in range(7):
pp.loglog([Dt[0], Dt[-1]],
[e0, e0 * (Dt[-1] / Dt[0]) ** o],
color='0.7')
pp.xlabel('dt')
pp.ylabel('||%s-%s_h||' % (label, label))
# pp.title('Method: %s' % method['name'])
ppl.legend(ax, loc=4)
pp.show()
return
def histogram(original, updated, bins=None, main="", save=None, log=False):
"""Plot a histogram of score improvements (updated-origianl)
Input:
original - list of original scores
updated - list of updates scores in same order as original
bins - number of bins to represent improvements
"""
#Lengths of score lists must be identical, assume in same order
assert len(original) == len(original)
#Set up bins:
if bins is not None and bins > 0:
imoprovements = {(-1,-1):0}
for i in xrange(0, len(original), bins):
improvements[(0,i+bins)] = 0
else:
improvements = {(-1,-1):0, (-5,0):0, (0,1):0, (1,25):0, (25,50):0, (50,75):0, (75,100):0, (100,125):0, (125,150):0, (150,200):0, (200,300):0, (300,400):0, (500,10000):0} #defaultdict(int)
#Calcualte improvements
for o, u in izip(original, updated):
if o>u:
improvements[(-1,-1)] += 1
continue
for lower, upper in improvements:
if lower <= int(u-o) < upper:
improvements[(lower,upper)] += 1
break
keys = sorted(improvements.keys(), key=lambda x:x[0])
values = [improvements[r] for r in keys]
fig, ax = plt.subplots()
ax.set_title(main)
ax.set_xlabel("Improvement (updated-original) bitscores")
ax.set_ylabel("log(Frequency)")
#ax.set_yscale('log')
width = 1.0
#ax.set_xticks(np.arange(len(improvements)))
#ax.set_xticklabels([l for l, u in keys])
bar(ax, np.arange(len(improvements)), values, log=log,
annotate=True, grid='y', xticklabels=[l for l, u in keys])
if save is None:
plt.show()
else:
plt.savefig(save)
def hero_stats_bar(self, thresh):
'''
Creates a bar chart of the heroes with highest
Wins / #Games played ratio
'''
# Create individual hero wins and losses
self.hero_stats()
fig, ax = plt.subplots(1, figsize=(9, 7))
# Compute the ratio of #Wins to the #Games played by that hero
# Most relevant statistic, better than W/L Ratio, better than
# just wins, all of them can be statistically insignificant
# in edge cases, but this can be the least of all
val = [
(k, self.wstat[k] / float(self.dstat[k] + self.wstat[k]))
for k in self.wstat
if self.wstat[k] / float(self.dstat[k] + self.wstat[k]) >= thresh
]
plt.title('Hero ID vs. Win Ratio (Matches from 01/23 - 02/24)')
plt.xlabel('Hero ID')
plt.ylabel('Win Ratio')
ax.set_xlim([0, len(val)])
ann = [round(k[1], 2) for k in val]
# Extract the xticklabels
xtl = [k[0] for k in val]
# Extract the individual values to be plotted
val = [k[1] for k in val]
ppl.bar(ax,
np.arange(len(val)),
val,
annotate=ann,
xticklabels=xtl,
grid='y',
color=ppl.colors.set2[2])
fig.savefig('../Figures/HIDvs#Wins#Games.png')
plt.show()
plt.clf()
xs,ys = np.where(np.isnan(sparse_eps))
for x,y in zip(xs,ys):
sparse_eps[x,y] = 0.5
max1 = np.max(sparse_eps)
max2 = np.max(dense_eps)
max3 = np.max(particle_eps)
maxtotal = np.max([max1,max2,max3])
min1 = np.min(sparse_eps)
min2 = np.min(dense_eps)
min3 = np.min(particle_eps)
mintotal = np.min([min1,min2,min3])
fig, (ax1,ax2,ax3) = ppl.subplots(3,1)
p1 = ppl.pcolormesh(fig,ax1,dense_eps)
p2 = ppl.pcolormesh(fig,ax2,sparse_eps)
p3 = ppl.pcolormesh(fig,ax3,particle_eps)
[p.set_clim(vmin=mintotal,vmax=maxtotal) for p in [p1,p2,p3]]
plt.show()
#print len(c)
for b in range(len(k)):
za = []
for a in range(len(e)):
if a != len(e) - 1:
za.append(c[b + len(k) * a])
z.append(za)
z = np.array(z)
#print z
#se consegue x e y (matrices) directamente conseguia
#X,Y = np.meshgrid(e,k)
# print z
fig, ax = plt.subplots()
cax = ax.pcolor(e, k, z, vmin=-25, vmax=175., cmap=('hot_r'))
plt.xscale('log')
plt.axis([1, 0.00223872113857, 2, 21])
plt.ylabel('<k>', rotation='vertical')
plt.xlabel(u'\u03B5')
plt.yticks([2, 4, 6, 8, 10, 12, 14, 16, 18, 20])
cbar = fig.colorbar(cax, ticks=[0, 25, 50, 75, 100, 125, 150, 175])
cbar.ax.set_yticklabels([0, 25, 50, 75, 100, 125, 150, 'N.C.'])
plt.show()
print 'Testing raw SVD => exact reconstruction'
svT = scipy.linalg.diagsvd(s, u.shape[0], vT.shape[1]).dot(vT)
for y in xrange(train.shape[0]):
for x in xrange(train.shape[1]):
colU = u[y, :]
rowV = svT[:, x]
assert np.allclose(train[y, x], single_dot(u, svT, x, y))
"""
##
plt.title('SVD reconstruction error on {}x{} matrix'.format(*train.shape))
plt.xlabel('Low rank approximation (k)')
plt.ylabel('Frobenius norm')
plt.ylim(0, max(svdY))
plt.legend(loc='best')
plt.savefig('reconstruct_fro_{}x{}.pdf'.format(*train.shape))
plt.show(block=True)
##
plt.plot(orthoX,
orthoY,
label="SVD",
color='black',
linewidth=2,
linestyle='--')
for label, X, Y in incr_ortho:
plt.plot(X, Y, label=label)
plt.title('SVD orthogonality error on {}x{} matrix'.format(*train.shape))
plt.xlabel('Low rank approximation (k)')
plt.ylabel('Deviation from orthogonality')
plt.semilogy()
#plt.ylim(0, max(orthoY))
plt.legend(loc='best')
np.random.shuffle(nodes)
print 'Pageviews from "real" edge weights'
print '-=-=-=-=-'
display_graph(G)
print
print 'Pageviews from evenly distributed edge weights'
print '-=-=-=-=-'
display_graph(approxG)
plt.plot(np.arange(0, len(rerrs)), rerrs, label='Relative error over time')
plt.xlabel('Iteration')
plt.ylabel('Average pageview relative error per node')
plt.legend()
plt.savefig('error_over_time.pdf')
plt.show(block=True)
plt.plot(np.arange(0, len(werrs)), werrs, label='Weight error over time')
plt.xlabel('Iteration')
plt.ylabel('Average weight error per edge')
plt.legend()
plt.savefig('weight_over_time.pdf')
plt.show(block=True)
fig, ax = plt.subplots(1)
ppl.bar(ax, *orig_weight_data, alpha=0.5, color='black', label='Weight error before')
ppl.bar(ax, np.arange(0, G.number_of_edges()), [abs(G[u][v]['weight'] - approxG[u][v]['weight']) for u, v in G.edges()], alpha=0.8, label='Weight error after')
#plt.ylim(-1, 1)
plt.legend(loc='best')
plt.show(block=True)
def scatter(original, updated, main="", save=None):
"""Plot a scatterplot of updated bitscores vs. original bitscores
"""
#Remove hits with no improvement and calcate the number of hits with no
#improvement(udated == original), positive imporvent (updated > original),
#and negative improvment (updated < original)
print len(original)
positiveImprovement = []
negativeImprovement = []
noImprovement = 0
for o, u in izip(original, updated):
if int(o) == int(u):
noImprovement +=1
elif u > o:
positiveImprovement.append((o,u))
elif u < o:
negativeImprovement.append((o,u))
else:
noImprovement +=1
if not positiveImprovement:
positiveImprovement = [()]
if not negativeImprovement:
negativeImprovement = [()]
print positiveImprovement
print negativeImprovement
print noImprovement
#Set deimensions
x, y = zip(*positiveImprovement+negativeImprovement)
xMax = int(round(sorted(x)[-1]/500.0)*500.0)
yMax = int(round(sorted(y)[-1]/500.0)*500.0)
sep = 500
xticks = range(0, xMax, sep)
yticks = range(0,yMax,sep)
color_cycle = brewer2mpl.get_map('Set2', 'qualitative', 8).mpl_colors
fig, ax = plt.subplots()
ax.set_title(main)
ax.set_xlabel("Original Bitscores")
ax.set_ylabel("Updated Bitscores")
#Plot postive improvement (green, automatically by prettyplotlib)
if positiveImprovement:
ppl.scatter(ax, *zip(*positiveImprovement),
label="Positive Improvement ({} seqs)".format(len(positiveImprovement)),
color=color_cycle[0])
#Draw no improvement line
ppl.plot(ax, (0,xMax), (0,xMax), color='k', linestyle='-', linewidth=2,
label="No Improvement ({} seqs)".format(noImprovement))
#Plot negative improvement (red, automatically by prettyplotlib)
if negativeImprovement:
ppl.scatter(ax, *zip(*negativeImprovement),
label="Negative Improvement ({} seqs)".format(len(negativeImprovement)),
color=color_cycle[1])
#Draw labels
ppl.legend(ax)
#Set axis
ax.set_ylim([0,yMax])
ax.set_xlim([0,xMax])
if save is None:
plt.show()
else:
pp = PdfPages(save)
pp.savefig(fig)
pp.close()
