def train(self, X, Y):
"""
Find the optimal split point
"""
# train
self.thresh = Threshold([x[0] for x in X],
[y == self.hit for y in Y])
def train(self, X, Y):
# construct values tables
table = {self.hit: [], self.miss: []}
for (x, y) in zip(X, Y):
table[y].append(x)
# transpose columns in table
for col in (self.hit, self.miss):
table[col] = zip(*table[col])
# compute weights
delta = mean(table[self.hit], 1) - mean(table[self.miss], 1)
covar = asmatrix(cov(table[self.hit]) + cov(table[self.miss]))
self.W = ravel(covar.I.dot(delta))
# compute threshold using this weight function
self.thresh = Threshold((self.score(x) for x in X), Y, self.hit)
class LDA(BinaryClassifier):
"""
Compute a linear discriminant classifier
>>> from csv import DictReader
>>> X = []
>>> Y = []
>>> for row in DictReader(open('iris.csv', 'r')):
... X.append([float(row['Sepal.Length']),
... float(row['Sepal.Width']),
... float(row['Petal.Length']),
... float(row['Petal.Width'])])
... Y.append(row['Species'])
>>> L = LDA(X, Y, 'versicolor')
>>> cm = L.leave_one_out(X, Y)
>>> round(cm.accuracy, 2)
0.96
>>> round(AUC(LDA, X, Y), 2)
1.0
"""
def __repr__(self):
return "{}(weights=[{}], {})".format(
self.__class__.__name__, ", ".join("{: 02.3f}".format(w) for w in self.W), self.thresh
)
def train(self, X, Y):
# construct values tables
table = {self.hit: [], self.miss: []}
for (x, y) in zip(X, Y):
table[y].append(x)
# transpose columns in table
for col in (self.hit, self.miss):
table[col] = zip(*table[col])
# compute weights
delta = mean(table[self.hit], 1) - mean(table[self.miss], 1)
covar = asmatrix(cov(table[self.hit]) + cov(table[self.miss]))
self.W = ravel(covar.I.dot(delta))
# compute threshold using this weight function
self.thresh = Threshold((self.score(x) for x in X), Y, self.hit)
def score(self, x):
return sum(w * f for w, f in zip(self.W, x))
def classify(self, x):
return self.hit if self.thresh.is_hit(self.score(x)) else self.miss
class Stump(BinaryClassifier):
"""
Compute a classifier which makes a single "cut" in a continuous
predictor vector X that splits the outcomes into "hit" and "miss"
so as to maximize the number of correct classifications
>>> from csv import DictReader
>>> X = []
>>> Y = []
>>> for row in DictReader(open('iris.csv', 'r')):
... X.append([float(row['Petal.Width'])])
... Y.append(row['Species'])
>>> s = Stump(X, Y, 'versicolor')
>>> s.leave_one_out(X, Y)
>>> round(s.accuracy(), 2)
0.88
>>> round(s.AUC(X, Y), 2)
0.99
"""
def __repr__(self):
lower = self.miss
upper = self.hit
if not self.thresh.hit_upper: # swap
(lower, upper) = (upper, lower)
return 'Stump({} < {: 02.3f} < {})'.format(lower,
self.thresh.split, upper)
def train(self, X, Y):
"""
Find the optimal split point
"""
# train
self.thresh = Threshold([x[0] for x in X],
[y == self.hit for y in Y])
def score(self, x):
return x[0]
def classify(self, x):
return self.hit if self.thresh.is_hit(self.score(x)) else self.miss
class Stump(BinaryClassifier):
"""
Compute a classifier which makes a single "cut" in a continuous
predictor vector X that splits the outcomes into "hit" and "miss"
so as to maximize the number of correct classifications
>>> from csv import DictReader
>>> X = []
>>> Y = []
>>> for row in DictReader(open('iris.csv', 'r')):
... X.append([float(row['Petal.Width'])])
... Y.append(row['Species'])
>>> s = Stump(X, Y, 'versicolor')
>>> s.leave_one_out(X, Y)
>>> round(s.accuracy(), 2)
0.88
>>> round(s.AUC(X, Y), 2)
0.99
"""
def __repr__(self):
lower = self.miss
upper = self.hit
if not self.thresh.hit_upper: # swap
(lower, upper) = (upper, lower)
return "Stump({} < {: 02.3f} < {})".format(lower, self.thresh.split, upper)
def train(self, X, Y):
"""
Find the optimal split point
"""
# train
self.thresh = Threshold([x[0] for x in X], [y == self.hit for y in Y])
def score(self, x):
return x[0]
def classify(self, x):
return self.hit if self.thresh.is_hit(self.score(x)) else self.miss
def train(self, X, Y):
"""
Find the optimal split point
"""
# train
self.thresh = Threshold([x[0] for x in X], [y == self.hit for y in Y])
