New Estimators

MANIFEST.in

@@ -1,7 +1,7 @@
 include *.rst
 recursive-include doc *
 recursive-include examples *
-recursive-include sliced/datasets *.csv
+recursive-include sliced/datasets *.csv *.tes *.tra
 include README.rst
 include requirements.txt
 include test_requirements.txt

examples/plot_iht.py

@@ -0,0 +1,27 @@
+"""
+================================
+Iterative Hessian Transformation
+================================
+
+An example plot of :class:`sliced.iht.IterativeHessianTransformation`
+applied to the Penn Digits Dataset.
+"""
+import numpy as np
+import matplotlib.pyplot as plt
+
+from sliced import IterativeHessianTransformation
+from sliced.datasets import load_penn_digits
+
+X_train, y_train = load_penn_digits(subset='train', digits=[0, 6, 9])
+X_test, y_test = load_penn_digits(subset='test', digits=[0, 6, 9])
+
+
+iht = IterativeHessianTransformation(target_type='response').fit(X_train, y_train)
+X_iht = iht.transform(X_test)
+
+# plot data projected onto the first direction
+plt.scatter(X_iht[:, 0], X_iht[:, 1], c=y_test, linewidth=0.5, edgecolor='k')
+plt.xlabel("$X\hat{\\beta_1}$")
+plt.ylabel("$X\hat{\\beta_2}$")
+
+plt.show()

examples/plot_kir.py

@@ -0,0 +1,36 @@
+"""
+=========================
+Kernel Inverse Regression
+=========================
+
+An example plot of :class:`sliced.kir.KernelInverseRegression`
+"""
+import numpy as np
+import matplotlib.pyplot as plt
+
+from sliced import KernelInverseRegression
+from sliced import datasets
+
+
+X, y = datasets.make_cubic(random_state=123)
+
+kir = KernelInverseRegression(gamma=1e-2)
+X_kir = kir.fit_transform(X, y)
+
+# estimate of the first dimension reducing directions
+beta1_hat = kir.directions_[0, :]
+
+
+# plot data projected onto the first direction
+plt.scatter(X_kir[:, 0], y, c=y, cmap='viridis', linewidth=0.5, edgecolor='k')
+plt.xlabel("$X\hat{\\beta_1}$")
+plt.ylabel("y")
+
+# annotation showing the direction found
+beta_text = "$\\beta_1$ = " + "{0}".format([0.707, 0.707])
+plt.annotate(beta_text, xy=(-2, 6.5))
+beta1_hat_text = "$\hat{\\beta_1}$ = " + "{0}".format(
+    np.round(beta1_hat, 3).tolist()[:2])
+plt.annotate(beta1_hat_text, xy=(-2, 7.5))
+
+plt.show()

examples/plot_opg.py

@@ -0,0 +1,35 @@
+"""
+======================
+Outer Product Gradient
+======================
+
+An example plot of :class:`sliced.odr.OuterProductGradient`
+"""
+import numpy as np
+import matplotlib.pyplot as plt
+
+from sliced import OuterProductGradients
+from sliced import datasets
+
+X, y = datasets.make_cubic(random_state=123)
+
+opg = OuterProductGradients()
+X_opg = opg.fit_transform(X, y)
+
+# estimate of the first dimension reducing directions
+beta1_hat = opg.directions_[0, :]
+
+
+# plot data projected onto the first direction
+plt.scatter(X_opg[:, 0], y, c=y, cmap='viridis', linewidth=0.5, edgecolor='k')
+plt.xlabel("$X\hat{\\beta_1}$")
+plt.ylabel("y")
+
+# annotation showing the direction found
+beta_text = "$\\beta_1$ = " + "{0}".format([0.707, 0.707])
+plt.annotate(beta_text, xy=(-2, 6.5))
+beta1_hat_text = "$\hat{\\beta_1}$ = " + "{0}".format(
+    np.round(beta1_hat, 3).tolist()[:2])
+plt.annotate(beta1_hat_text, xy=(-2, 7.5))
+
+plt.show()

examples/plot_phd.py

@@ -0,0 +1,27 @@
+"""
+==================================
+Principal Hessian Directions (PHD)
+==================================
+
+An example plot of :class:`sliced.phd.PrincipalHessianDirections`
+"""
+import numpy as np
+import matplotlib.pyplot as plt
+
+from sliced import PrincipalHessianDirections
+from sliced import datasets
+
+
+from sliced.datasets import load_athletes
+
+X, y = load_athletes()
+
+phd = PrincipalHessianDirections(target_type='residual')
+X_phd = phd.fit_transform(X, y)
+
+# plot data projected onto the first direction
+plt.scatter(X_phd[:, 0], y, c=y, cmap='viridis', linewidth=0.5, edgecolor='k')
+plt.xlabel("$X\hat{\\beta_1}$")
+plt.ylabel("y")
+
+plt.show()

sliced/__init__.py

@@ -8,5 +8,12 @@
 from sliced import externals
 
 from sliced.sir import *
+from sliced.kir import *
 from sliced.save import *
+from sliced.phd import *
+from sliced.iht import *
+from sliced.opg import *
+from sliced.dr import *
+from sliced.ladle import *
+from sliced.bic import *
 from sliced.version import *

sliced/bic.py

@@ -0,0 +1,48 @@
+import numpy as np
+
+from sklearn.base import clone, BaseEstimator, MetaEstimatorMixin
+from sklearn.base import TransformerMixin
+from sklearn.utils import resample, check_random_state, check_array
+
+
+__all__ = ['BICDimensionSelector']
+
+
+class BICDimensionSelector(MetaEstimatorMixin, BaseEstimator, TransformerMixin):
+    def __init__(self, estimator, criterion='log'):
+        self.estimator = estimator
+        self.criterion = criterion
+
+    def fit(self, X, y):
+        n_samples, n_features = X.shape
+
+        # fit sufficient dimension reduction method
+        self.estimator_ = clone(self.estimator)
+        self.estimator_.fit(X, y)
+
+        evals = self.estimator_.eigenvalues_
+        self.criterion_ = np.zeros(n_features + 1)
+        if self.criterion == 'sum':
+            penalty = 2 * evals[0] * ((n_samples)**(-0.5)) * np.log(n_samples)
+            for k in range(n_features + 1):
+                if k == 0:
+                    self.criterion_[k] = 0
+                else:
+                    self.criterion_[k] = np.sum(evals[:k])
+                    self.criterion_[k] -= penalty * k
+        elif self.criterion == 'log':
+            penalty = (evals[0] / 3.) * (np.log(n_samples) + np.sqrt(n_samples))
+            penalty /= (2. * n_samples)
+            for k in range(n_features + 1):
+                if k != n_features:
+                    self.criterion_[k] = np.sum(np.log(evals[k:] + 1) - evals[k:])
+                self.criterion_[k] -= penalty * k * (2 * n_features - k + 1)
+        else:
+            raise ValueError('Unrecognized criterion {}'.format(self.criterion))
+
+        self.n_directions_ = np.argmax(self.criterion_)
+
+        return self
+
+    def transform(self, X):
+        return self.estimator_.transform(X)

sliced/datasets/__init__.py

@@ -1,3 +1,4 @@
 from sliced.datasets.synthetic import *
 from sliced.datasets.banknote import load_banknote
 from sliced.datasets.athletes import load_athletes
+from sliced.datasets.penn_digits import load_penn_digits