def findErrorBarsBindingCurve(subSeries):
""" Return bootstrapped confidence intervals on columns of an input data matrix.
Assuming rows represent replicate measurments, i.e. clusters. """
eminus=[]
eplus = []
for i in subSeries:
vec = subSeries.loc[:, i].dropna()
success = True
if len(vec) > 1:
try:
bounds = bootstrap.ci(vec, np.median, n_samples=1000)
except IndexError:
success = False
else:
success = False
if success:
eminus.append(vec.median() - bounds[0])
eplus.append(bounds[1] - vec.median())
else:
eminus.append(np.nan)
eplus.append(np.nan)
eminus = pd.Series(eminus, index=subSeries.columns)
eplus = pd.Series(eplus, index=subSeries.columns)
return eminus, eplus
def returnFractionGroupedBy(mat, param_in, param_out):
""" Given data matrix, return mean of param_out binned by param_in. """
grouped = mat.groupby(param_in)[param_out]
y = grouped.mean()
x = y.index.tolist()
with warnings.catch_warnings():
warnings.simplefilter("ignore")
yerr = np.array([np.abs(bootstrap.ci(group, method='pi', n_samples=100) - y.loc[name])
for name, group in grouped]).transpose()
return x, y, yerr
def findAlpha(self, n):
x = self.x
y = self.y
index = (x>0)&(y>0)
#vec = np.exp(np.log(y/x)/n).loc[index]
vec = np.power(y/x, 1/float(n)).loc[index]
alpha = vec.median()
lb, ub = bootstrap.ci(vec, statfunction=np.median, n_samples=1000)
xlim = [0.9 ,1.1]
bins = np.arange(0.898, 1.1, 0.001)
plt.figure(figsize=(4,3));
sns.distplot(vec, bins=bins, hist_kws={'histtype':'stepfilled'}, kde_kws={'clip':xlim});
plt.axvline(alpha, color='k', linewidth=0.5);
plt.axvline(lb, color='k', linestyle=':', linewidth=0.5);
plt.axvline(ub, color='k', linestyle=':', linewidth=0.5);
plt.xlabel('photobleach fraction per image');
plt.ylabel('probability');
plt.xlim(xlim)
fix_axes(plt.gca());
plt.tight_layout()
return alpha, lb, ub
'substrate']].dropna()
merge.rename(columns={'substrate_x': 'substrate'}, inplace=True)
del df1, df2, df3
grouped = merge.groupby('substrate')
sand = grouped.get_group('sand')
gravel = grouped.get_group('gravel')
boulders = grouped.get_group('boulders')
del merge
print 'Calculating calibrations metrics for lsq classifications...'
calib_df = pd.DataFrame(columns=['ent', 'h**o', 'var'],
index=['sand', 'gravel', 'boulders'])
calib_df.loc['sand'] = pd.Series({
'h**o':
1 - np.average(boot.ci(sand['homo_median'], np.median, alpha=0.05)),
'ent':
np.average(boot.ci(sand['entropy_median'], np.median, alpha=0.05)),
'var':
np.average(boot.ci(sand['var_median'], np.median, alpha=0.05))
})
calib_df.loc['gravel'] = pd.Series({
'h**o':
1 - np.average(boot.ci(gravel['homo_median'], np.median, alpha=0.05)),
'ent':
np.average(boot.ci(gravel['entropy_median'], np.median, alpha=0.05)),
'var':
np.average(boot.ci(gravel['var_median'], np.median, alpha=0.05))
})
calib_df.loc['boulders'] = pd.Series({
'h**o':
'std': 'var_std'
},
inplace=True)
merge = df1.merge(df2, left_index=True, right_index=True, how='left')
merge = merge.merge(df3, left_index=True, right_index=True, how='left')
merge = merge[['homo_median', 'entropy_median', 'var_median',
'substrate']].dropna()
merge.rename(columns={'substrate_x': 'substrate'}, inplace=True)
del df1, df2, df3
grouped = merge.groupby('substrate')
sand = grouped.get_group('sand')
gravel = grouped.get_group('gravel')
boulders = grouped.get_group('boulders')
boot.ci(sand['homo_median'], np.median, alpha=0.05)
boot.ci(sand['entropy_median'], np.median, alpha=0.05)
boot.ci(sand['var_median'], np.median, alpha=0.05)
boot.ci(gravel['homo_median'], np.median, alpha=0.05)
boot.ci(gravel['entropy_median'], np.median, alpha=0.05)
boot.ci(gravel['var_median'], np.median, alpha=0.05)
boot.ci(boulders['homo_median'], np.median, alpha=0.05)
boot.ci(boulders['entropy_median'], np.median, alpha=0.05)
boot.ci(boulders['var_median'], np.median, alpha=0.05)
homo_df = pd.DataFrame(index=['Sand', 'Gravel', 'Boulders'],
columns=['lbound', 'ubound'])
homo_df.loc['Sand'] = pd.Series({
'lbound':
def get_final_ci(series):
mean = series.mean()
std = series.std(ddof=1)
low_ci, high_ci = bootstrap.ci(series, n_samples=1000)
return mean, std, low_ci, high_ci
