def errors (shape, radius = 20.0, trials = 100):
'Plot shape fitting errors for multiple MC trials'
errors = []
for i in range(trials):
errors.append(fit(shape, radius))
def safe (e):
e = abs(e)
if e < 1:
return e
else:
return 1
r_errors = array([safe(e[0]) for e in errors])
xy_errors = array([safe(sqrt(e[2]**2 + e[2]**2)) for e in errors])
r_mean = exp(sum(log(r_errors)) / trials)
xy_mean = exp(sum(log(xy_errors)) / trials)
if main.at_top():
pyplot.figure()
pyplot.loglog(r_errors, xy_errors, 'r.')
pyplot.loglog([r_mean],[xy_mean], 'k+')
pyplot.xlabel('radius error (px)')
pyplot.ylabel('position error (px)')
pyplot.title('Precision of %g pixel %s markers (%i trials)'
% (2 * radius, shape.capitalize(), trials))
pyplot.savefig('%s-%g-error.png' % (shape, radius))
pyplot.savefig('%s-%g-error.pdf' % (shape, radius))
pyplot.show()
return r_mean,xy_mean
def search (self, query):
self.add_new_results(query)
# return new + old results
results = list(self.db_cursor.execute(
'''
select
tweets.tweet_id,
tweets.from_user_id,
tweets.to_user_id,
tweets.geo,
tweets.created_at,
tweets.text
from search_results join tweets
on search_results.tweet_id = tweets.tweet_id
where query = (?)
order by tweets.tweet_id
''',
(query,)))
if main.at_top():
for result in results:
text = results[-1]
print '\n%s' % text
else:
return results
def layers (offset = 4, *positions):
'Plots grid with offset positions'
positions = set(int(p) for p in positions)
positions.add(0)
radius = 3 * 12
R = range(radius+1)
def marked (p,x,y):
return (x + offset * y) % 12 == p
pyplot.figure()
#pyplot.axis('equal')
pyplot.title('offset = %i, positions = %s' % (offset, list(positions)))
for r in R:
pyplot.plot([0,radius],[r,r], color = (0.9,)*3)
pyplot.plot([r,r],[0,radius], color = (0.9,)*3)
for p in positions:
XY = [(x,y) for x in R for y in R if marked(p,x,y)]
X = [x for x,y in XY]
Y = [y for x,y in XY]
pyplot.plot(X,Y, marker = '.', linestyle = 'none')
pyplot.xlim(0,radius)
pyplot.ylim(0,radius)
if main.at_top():
pyplot.show()
def search (query):
'searches twitter & caches result in local database'
print query
results = searcher.search(query)
if main.at_top():
for t in results:
print '-' * 80
print t
print '\nsearch returned %i results' % len(results)
else:
return results
def ranges (shape, min_rad = 1.0, max_rad = 20.0, steps = 100):
'Plot sampled fitting error vs radius'
L0 = log(min_rad)
L1 = log(max_rad)
dL = (L1 - L0) / steps
radii = exp(arange(L0, L1, dL))
r_errors = []
xy_errors = []
for radius in radii:
r,xy = errors(shape, radius, trials = 1)
r_errors.append(r)
xy_errors.append(xy)
r_errors = array(r_errors)
xy_errors = array(xy_errors)
if main.at_top():
pyplot.figure()
pyplot.loglog(2 * radii, 2 * r_errors, 'k.')
pyplot.xlabel('marker size (px)')
pyplot.ylabel('size error (px)')
pyplot.ylim(1e-3, 1)
pyplot.title('%s marker radius precision' % shape.capitalize())
pyplot.savefig('%s-radius-ranges.png' % shape)
pyplot.savefig('%s-radius-ranges.pdf' % shape)
pyplot.figure()
pyplot.loglog(2 * radii, xy_errors, 'k.')
pyplot.xlabel('marker size (px)')
pyplot.ylabel('position error (px)')
pyplot.ylim(1e-3, 1)
pyplot.title('%s marker position precision' % shape.capitalize())
pyplot.savefig('%s-position-ranges.png' % shape)
pyplot.savefig('%s-position-ranges.pdf' % shape)
pyplot.show()
return radii, r_errors, xy_errors
def curves (*radii):
'Plot geometric mean fitting error vs radius'
radii = [float(r) for r in radii]
radii.sort()
r_shapes = dict((shape,[]) for shape in shapes)
xy_shapes = dict((shape,[]) for shape in shapes)
for radius in radii:
for shape in shapes:
r,xy = errors(shape, radius)
r_shapes[shape].append(r)
xy_shapes[shape].append(xy)
if main.at_top():
pyplot.figure()
for s in shapes:
pyplot.plot(radii, r_shapes[s], label=s)
pyplot.xlabel('marker radius (px)')
pyplot.ylabel('mean radius error (px)')
pyplot.title('Precision of shapes: %s' % ', '.join(shapes))
pyplot.legend()
pyplot.savefig('radius-curves.png')
pyplot.savefig('radius-curves.pdf')
pyplot.figure()
for s in shapes:
pyplot.plot(radii, xy_shapes[s], label=s)
pyplot.xlabel('marker radius (px)')
pyplot.ylabel('position error (px, geometric mean)')
pyplot.title('Precision of shapes: %s' % ', '.join(shapes))
pyplot.legend()
pyplot.savefig('position-curves.png')
pyplot.savefig('position-curves.pdf')
pyplot.show()
def fit (shape, radius = 20.0, iters = 10):
'Fit shape to noisy image, display pixel errors'
draw = draw_shape[shape]
size = int(3 * radius)
im = draw(size, radius, 0.5 * size, 0.5 * size)
im += 2 * noise_level * randn(*im.shape)
truth = array([
log(radius),
0.5 * size,
0.5 * size,
])
sigma = array([
0.1,
1.0,
1.0,
])
mean = truth + sigma * randn(3)
sigma
fun = drawing_error(draw,im)
def print_mean ():
r,x,y = exp(mean[0]),mean[1],mean[2]
print 'radius = %g, x = %g, y = %g' % (r,x,y)
print 'fitting dot radius of %i x %i image' % im.shape
print_mean()
optim.nonlinear_minimize(fun, mean, sigma, iters)
print_mean()
if main.at_top():
fun(mean, True)
error = mean - truth
print 'error = %s' % error
return error
