pimaclassifier.py

def accuracy(truth, predicted):
    if len(truth) != len(predicted):
        raise Exception("Wrong sizes ...")
    total = len(truth)
    if total == 0: 
        return 0  
    hits = len(filter(lambda (x, y): x == y, zip(truth, predicted)))
    return float(hits)/total


import pandas as pd
Pima_tr = pd.read_csv("Pima.tr.csv", index_col=0)
Pima_te = pd.read_csv("Pima.te.csv", index_col=0)
Pima_tr.groupby("type").count()


class NoClassifier():
    """Classifier that predict all data as "No". """
    def predict(self, x):
        return pd.Series(["No"] * x.shape[0])
          

no_classifier = NoClassifier()
predicted = no_classifier.predict(Pima_te)
accuracy(Pima_te.type, predicted) 


from scipy.linalg import pinv
from numpy import asarray, hstack, mat, ones, where

class LinearClassifier():
    """ y = X*b and b = pinv(X) * y """
    def __init__(self): 
        self._parameters = None
    
    def from_labels(self, y):
        return mat(where(y=="No", -1, 1)).T
    
    def to_labels(self, y):
        return pd.Series(asarray(where(y<0, "No", "Yes")).flatten())
    
    def train(self, x, y):
        intercept = ones((x.shape[0], 1))
        self._parameters = pinv(hstack((mat(x), intercept))) * self.from_labels(y)
    
    def predict(self, x):
        intercept = ones((x.shape[0], 1))
        y_estimated = hstack((mat(x), intercept)) * self._parameters
        return self.to_labels(y_estimated)



lc = LinearClassifier()
lc.train(Pima_tr.ix[:,:7], Pima_tr.type)
predicted = lc.predict(Pima_te.ix[:,:7])
accuracy(Pima_te.type, predicted)


from numpy import where

X_tr = Pima_tr.ix[:,:7]
y_tr = where(Pima_tr.type=="No", -1, 1)
X_te = Pima_te.ix[:,:7]
y_te = where(Pima_te.type=="No", -1, 1)

from sklearn.ensemble import RandomForestClassifier
rfc = RandomForestClassifier()
rfc.fit(X_tr, y_tr)
predicted = where(rfc.predict(X_te)==-1, "No", "Yes")
accuracy(Pima_te.type, predicted)
rfc.score(X_te, y_te)


# Modified slightly from http://scikit-learn.org/stable/auto_examples/plot_roc.html
from sklearn.metrics import roc_curve, auc
import matplotlib.pyplot as plt

fpr, tpr, thresholds = roc_curve(y_te, rfc.predict_proba(X_te)[:,1])
roc_auc = auc(fpr, tpr)

plt.plot(fpr, tpr, label='ROC curve (area = %0.2f)' % roc_auc)
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.legend(loc="lower right")
plt.show()