def test_ace_multi_targets_bg_eq_zero(self):
'''ACE score of background for multiple targets should be one.'''
ij1 = (10, 10)
ij2 = (3, 12)
y = spy.ace(self.bg.mean, [self.X[ij1], self.X[ij2]],
background=self.bg)
assert (np.allclose(0, y))
def test_ace_subspace_bg_eq_zero(self):
'''ACE score of background for target subspace should be zero.'''
ij1 = (10, 10)
ij2 = (3, 12)
y = spy.ace(self.bg.mean, np.array([self.X[ij1], self.X[ij2]]),
background=self.bg)
assert(np.allclose(0, y))
def test_ace_multi_targets_bg_eq_zero(self):
'''ACE score of background for multiple targets should be one.'''
ij1 = (10, 10)
ij2 = (3, 12)
y = spy.ace(self.bg.mean, [self.X[ij1], self.X[ij2]],
background=self.bg)
assert(np.allclose(0, y))
def test_ace_novec_subspace_targets_eq_one(self):
'''ACE score of targets defining target subspace should each be one.'''
ij1 = (10, 10)
ij2 = (3, 12)
y = spy.ace(self.X, np.array([self.X[ij1], self.X[ij2]]),
background=self.bg, vectorize=False)
assert(np.allclose(1, [y[ij1], y[ij2]]))
def test_ace_novec_multi_targets_eq_one(self):
'''ACE score of multiple targets should each be one.'''
ij1 = (10, 10)
ij2 = (3, 12)
y = spy.ace(self.X, [self.X[ij1], self.X[ij2]], background=self.bg,
vectorize=False)
assert(np.allclose(1, [y[ij1][0], y[ij2][1]]))
def test_ace_novec_pixel_target_eq_one(self):
'''ACE score of target should be one for single pixel arg.'''
ij = (10, 10)
y = spy.ace(self.X[ij],
self.X[ij],
background=self.bg,
vectorize=False)
assert (np.allclose(1, y))
def cross_validate():
#for tinkering with the model
#read data
print('loading data')
all_df = pd.read_csv('./data/train.csv',index_col = 'ID')
#feature selection
#feature_selection = feature_importance(all_df) > 0.0001
#feature_selection = list(feature_selection.ravel())
#feature_selection.append(True)
#all_df = all_df[all_df.columns[feature_selection]]
#split data into training and cross validation set
zeros_df = all_df[all_df.TARGET == 0]
ones_df = all_df[all_df.TARGET == 1]
num_ones = ones_df.shape[0]
msk = np.random.rand(len(zeros_df)) < 0.1*num_ones/len(zeros_df)
zeros_train_df = zeros_df[~msk]
zeros_test_df = zeros_df[msk]
msk = np.random.rand(num_ones) < 0.1
ones_train_df = ones_df[~msk]
ones_test_df = ones_df[msk]
train_df = pd.concat([zeros_train_df,ones_train_df,zeros_test_df])
targets_df = ones_test_df
#test_df has approx 50% 1s and 50% 0s
#oversample with smote
train_X = np.array(train_df.drop('TARGET',axis = 1))
train_Y = np.array(train_df.TARGET)
#predict on smoted data
print('predicting')
targets = np.array(targets_df.drop('TARGET',axis = 1))
output = ace(train_X,targets)
print(output)
#score
print('scoring')
conf_matrix = confusion_matrix(test_Y,predictions)
print('confusion matrix:')
print(pd.DataFrame(conf_matrix,columns = [0,1]))
print('accuracy:')
print(sum(test_Y.reshape(predictions.shape) == predictions)/len(test_Y))
def test_ace_windowed_target_eq_one(self):
'''ACE score of target for windowed background should be one.'''
ij = (10, 10)
y = spy.ace(self.X, self.X[ij], window=(3, 7), cov=self.bg.cov)
assert (np.allclose(1, y[ij]))
def test_ace_novec_target_eq_one(self):
'''ACE score (without vectorization) of target should be one.'''
ij = (10, 10)
y = spy.ace(self.X, self.X[ij], background=self.bg, vectorize=False)
assert (np.allclose(1, y[ij]))
def test_ace_target_eq_one(self):
'''ACE score of target should be one.'''
ij = (10, 10)
y = spy.ace(self.X, self.X[ij], background=self.bg)
assert (np.allclose(1, y[ij]))
def test_ace_bg_eq_zero(self):
'''ACE score of background mean should be zero.'''
ij = (10, 10)
y = spy.ace(self.bg.mean, self.X[ij], background=self.bg)
assert (np.allclose(0, y))
def test_ace_windowed_target_eq_one(self):
'''ACE score of target for windowed background should be one.'''
ij = (10, 10)
y = spy.ace(self.X, self.X[ij], window=(3,7), cov=self.bg.cov)
assert(np.allclose(1, y[ij]))
def test_ace_novec_target_eq_one(self):
'''ACE score (without vectorization) of target should be one.'''
ij = (10, 10)
y = spy.ace(self.X, self.X[ij], background=self.bg, vectorize=False)
assert(np.allclose(1, y[ij]))
def test_ace_target_eq_one(self):
'''ACE score of target should be one.'''
ij = (10, 10)
y = spy.ace(self.X, self.X[ij], background=self.bg)
assert(np.allclose(1, y[ij]))
def test_ace_novec_pixel_target_eq_one(self):
'''ACE score of target should be one for single pixel arg.'''
ij = (10, 10)
y = spy.ace(self.X[ij], self.X[ij], background=self.bg, vectorize=False)
assert(np.allclose(1, y))
def test_ace_bg_eq_zero(self):
'''ACE score of background mean should be zero.'''
ij = (10, 10)
y = spy.ace(self.bg.mean, self.X[ij], background=self.bg)
assert(np.allclose(0, y))
