def plot():
plot = ctn_benchmark.Plot([
ctn_benchmark.Data('nengo'),
ctn_benchmark.Data('ocl'),
ctn_benchmark.Data('spinn'),
ctn_benchmark.Data('sp_rmv')
])
plot.measures(['period', 'period_sd', 'peak', 'peak_sd'],
show_outliers=False)
import pylab
pylab.show()
def fit():
data = ctn_benchmark.Data('data')
n_neurons = 200
for d in data.data[:]:
if d['_n_neurons'] != n_neurons:
data.data.remove(d)
target_data = [0.97, 0.94, 0.91, 0.80] # experimental data
import scipy.stats
def prob(x, noise=1.0):
return scipy.stats.norm.cdf(np.array(x) / noise)
def curve(x, noise):
return np.mean([prob(d['values'], noise=noise) for d in data.data],
axis=0)
import scipy.optimize
p, err = scipy.optimize.curve_fit(curve, np.arange(3), target_data)
import pylab
pylab.plot([2, 4, 6, 8], target_data)
pylab.plot([2, 4, 6, 8], curve(None, *p))
pylab.show()
print p, err
def plot():
data = ctn_benchmark.Data('data')
plt = ctn_benchmark.Plot(data)
plt.lines('_n_neurons', ['rmse'])
import pylab
pylab.show()
def plot():
data = ctn_benchmark.Data('data')
plot = ctn_benchmark.Plot(data)
plot.vary('_N_reference', ['score1', 'score2', 'score3', 'score4'])
import pylab
pylab.show()
def gather():
data = ctn_benchmark.Data('compare')
d = []
for i in range(1, 9):
d.append(1.0 - data.get('score%d' % i))
d = np.array(d).T
for row in d:
print ', '.join(['%1.3f' % x for x in row])
def plot():
data = ctn_benchmark.Data('compare')
plot = ctn_benchmark.Plot(data)
plot.measures([
'score1', 'score2', 'score3', 'score4', 'score5', 'score6',
'score7', 'score8'
],
outlier_cutoff=None,
ylim=(0, 1))
import pylab
pylab.show()
def plot():
data = ctn_benchmark.Data('data')
import pylab
pylab.figure(figsize=(10, 5))
Ns = np.linspace(10, 200, 5)
for i, N in enumerate(Ns):
pylab.subplot(1, len(Ns), i + 1)
pylab.title('N=%d' % N)
sd = 20
N_refs = data.get('_N_reference')
weights = np.exp(-(N_refs - N)**2 / (2 * sd**2))
weights /= np.sum(weights)
measures = ['score1', 'score2', 'score3', 'score4']
low = []
high = []
sample = []
for m in measures:
d = data.get(m)
d = 1.0 - d
ci = ctn_benchmark.stats.bootstrapci(d,
np.mean,
weights=weights)
low.append(ci[0])
high.append(ci[1])
sample.append(np.average(d, weights=weights))
x = np.arange(len(measures)) + 1
pylab.fill_between(x, low, high, color='#aaaaaa')
pylab.plot(x, sample)
pylab.ylim(0, 1)
pylab.xticks(x)
pylab.xlim(0.5, len(measures) + 0.5)
pylab.tight_layout()
pylab.show()