Skip to main content

Feature selection with FDR control.

Project description

>>> import numpy as np
>>> from fs_fdr import barber_candes_selection, knockoff_features_construction, utils
>>>
>>> from sklearn.ensemble import GradientBoostingClassifier
>>> #######################################################################################
>>> #### Simulate data to check the performance of the methods.
>>> #######################################################################################
>>>
>>>
>>> type = "classification"
>>> # First simulate some data
>>> n, p, p1 = 1000, 50, 20
>>> rho = 0.
>>> mean = 0.
>>> sd = 1.
>>> error_std = 1.
>>> r = ["uniform", .0, .5]
>>> x = mean + sd * np.random.normal(0., 1., size=(n, p))
>>> true_w = np.random.uniform(r[1], r[2], size=(p1, 1))
>>> negate = np.random.binomial(n=1, p=.5, size=(p1, 1))
>>> negate[np.where(negate==0.), :] = -1
>>> true_w = true_w * negate
>>> true_index = np.random.choice(np.arange(p), size = p1, replace=False)
>>> true_index = np.sort(true_index)
>>> xbeta = np.dot(x[:, true_index], true_w)
>>> pr = 1/(1+np.exp(-xbeta))
>>> t = (pr > .5) + 0.
>>>
>>> q = .1
>>> # Step 2: Create knockoff features using the knockoff_method library
>>>
>>> # Set parameters
>>>
>>> selection_method = "knockoff-MX"
>>> optimization = ["ASDP", "selfblocks", 50, 50]
>>>
>>> VI_stat = "Diff"
>>>
>>>
>>> myknockoff = knockoff_features_construction.Knockoff(x, selection_method, optimization)
>>> knockoff_attrs = myknockoff.knockoff_features()
>>> x, x_tilda = knockoff_attrs.X, knockoff_attrs.X_tilde
>>>
>>>
>>>
>>> modeling = {"model":"gradient boosting", "params":"classification fi"}
>>>
>>>
>>> data = [x, x_tilda, t]
>>> knockoff_selection = barber_candes_selection.BarberCandesSelection(data, modeling, selection_method,q=q, VI_stat=VI_stat).selection()
>>>
>>> S_knock = knockoff_selection.S
>>> FDR_UB = knockoff_selection.FDR_UB
>>>
>>>
>>>
>>> fdr_knock = 100*utils.FDR(S_knock, true_index)
>>> power_knock = 100*utils.power(S_knock, true_index)
>>> fnp_knock = 100*utils.FNP(S_knock, true_index, p)
>>> print('------------Knockoff ({})-------------'.format(modeling["model"]))
>>> print("Empirical FDR: " + str(100*np.round(FDR_UB, 2)) + "%")
>>> print("FDR:  " +str(fdr_knock) + "%")
>>> print("power:  "+str(power_knock) + "%")
>>> print("FNP:  "+str(fnp_knock) + "%")
>>>
>>>
>>> ##########DSS
>>>
>>>
>>> modeling = {"model":"gradient boosting", "params":"classification fi"}
>>> selection_method = "DSS"
>>> data = [x, t]
>>> split_type = ["sampling", 5, 5]
>>> prob = .7
>>> DSS_selection = barber_candes_selection.BarberCandesSelection(data, modeling, selection_method,q=q).selection()
>>>
>>> S_dss = DSS_selection.S
>>> FDR_UB = DSS_selection.FDR_UB
>>>
>>>
>>>
>>> fdr_dss = 100*utils.FDR(S_dss, true_index)
>>> power_dss = 100*utils.power(S_dss, true_index)
>>> fnp_dss = 100*utils.FNP(S_dss, true_index, p)
>>> print('------------DSS ({})-------------'.format(modeling["model"]))
>>> print("Empirical FDR: " + str(100*np.round(FDR_UB, 2)) + "%")
>>> print("FDR:  " +str(fdr_dss) + "%")
>>> print("power:  "+str(power_dss) + "%")
>>> print("FNP:  "+str(fnp_dss) + "%")
>>>
>>>
>>>
>>> ############### SVM
>>>
>>> modeling = {"model": "not specified", "params":"given"}
>>>
>>> from sklearn.svm import SVC
>>>
>>> svm = SVC(C=1., kernel="linear")
>>> svm_w = svm.fit(np.hstack((x, x_tilda)), t.ravel()).coef_
>>>
>>> selection_method = "knockoff"
>>> knockoff_selection = barber_candes_selection.BarberCandesSelection(modeling=modeling, selection_method=selection_method, w = svm_w).selection()
>>>
>>> S_knock = knockoff_selection.S
>>> FDR_UB = knockoff_selection.FDR_UB
>>> print("empirical FDR: " + str(100*np.round(FDR_UB, 2)))
>>>
>>> fdr_knock = 100*utils.FDR(S_knock, true_index)
>>> power_knock = 100*utils.power(S_knock, true_index)
>>> fnp_knock = 100*utils.FNP(S_knock, true_index, p)
>>> print('------------Knockoff ({})-------------'.format(modeling["model"]))
>>> print("Empirical FDR: " + str(100*np.round(FDR_UB, 2)) + "%")
>>> print("FDR:  " +str(fdr_knock) + "%")
>>> print("power:  "+str(power_knock) + "%")
>>> print("FNP:  "+str(fnp_knock) + "%")

Project details


Release history Release notifications | RSS feed

This version

0.1

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

feature_selection_fdr-0.1.tar.gz (18.8 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page