gistfile1.py

# make inverse fuction for SciPy's ECDF

import numpy as np
import statsmodels as sm
import scipy as sp

num_samples = 1000
values = np.random.randn(num_samples)
ecdf = sm.distributions.ECDF(values)
ecdf_i = sp.interpolate.interp1d(ecdf.y,ecdf.x)

quant1 = 0.7
val1 = np.percentile(values,quant1*100)
quant2 = ecdf(val1)
val2 = ecdf_i(quant2)

print "quantile %s ~ value %s ~ quantile %s ~ value %s\n"%(quant1,val1,quant2,val2)
print "quantile error: %s"%(quant2-quant1)
print "value error: %s"%(val1-val2)

### sample output:
# quantile 0.7 ~ value 0.507576803872 ~ quantile 0.7 ~ value 0.507447104053
#
# quantile error: 1.11022302463e-16
# value error: 0.000129699818594