def test_f_stat_advertising(self):
'''
This is testing f-stat for using data from ISLR pg. 76 to check against.
'''
df = pd.read_csv('Advertising.csv')
# Set Y to sales
Y = df['sales']
# Set X to newspaper
X = pd.DataFrame(df[['TV', 'radio', 'newspaper']])
# Create a linear model and train
reg = LinearRegression()
reg.fit(X, Y)
# Get predictions for Y and then calculate items for t-stat
Y_pred = reg.predict(X)
tss = lms.tss(Y)
rss = lms.rss(Y.values, Y_pred)
f_stat = lms.f_stat(tss, rss, Y.size, X.columns.size)
actual = round(f_stat)
expected = 570
self.assertEqual(actual, expected)
def test_p_value_advertising_full_X(self):
'''
This is testing p-value for newspaper using data from ISLR pg. 74 to check against
'''
places = 5
df = pd.read_csv('Advertising.csv')
# Set Y to sales
Y = df['sales']
# Set X to newspaper
X = pd.DataFrame(df[['TV', 'radio', 'newspaper']])
# Create a linear model and train
reg = LinearRegression()
reg.fit(X, Y)
# Get predictions for Y and then calculate rse from this.
Y_pred = reg.predict(X)
rss = lms.rss(Y.values, Y_pred)
rse = lms.rse(rss, Y.values.size, X.columns.size)
var = rse**2
standard_error = lms.standard_error(var, X['newspaper'].values)
t_stat = lms.t_stat(reg.coef_[2], standard_error)
actual = standard_error
expected = -0.0059
self.assertAlmostEqual(actual, expected, places=4)
def test_t_stat_advertising(self):
'''
This is testing standard error using data from ISLR pg. 68 and pg. 72 to check against.
'''
t_stat_data = [('TV', 17.67), ('radio', 9.92), ('newspaper', 3.30)]
places = 2
df = pd.read_csv('Advertising.csv')
# Set Y to sales
Y = df['sales']
# Iterate over standard error data from ISLR and test each
for i in range(len(t_stat_data)):
t_stat_tup = t_stat_data[i]
# Set X to column
X = pd.DataFrame(df[t_stat_tup[0]])
# Create a linear model and train
reg = LinearRegression()
reg.fit(X, Y)
# Get predictions for Y and then calculate rse from this.
Y_pred = reg.predict(X)
rss = lms.rss(Y.values, Y_pred)
rse = lms.rse(rss, Y.values.size, X.columns.size)
var = rse**2
standard_error = lms.standard_error(var, X[t_stat_tup[0]].values)
actual = lms.t_stat(reg.coef_[0], standard_error)
expected = t_stat_tup[1]
self.assertAlmostEqual(actual, expected, places=places)
def test_rss_basic(self):
'''
This is just a basic test with a small amount of variables that can be easily calculated.
'''
Y1 = np.array([1, 3, 7])
Y2 = np.array([0, 6, 9])
actual = lms.rss(Y1, Y2)
expected = 14
self.assertEqual(actual, expected)
def test_rss_len(self):
'''
This test calculates RSS in a way where RSS can be calculated as a a factor of the length.
'''
Y1 = np.arange(100)
for i in range(1, 15):
Y2 = Y1 + i
fac = i**2
actual = lms.rss(Y1, Y2)
expected = Y1.size * fac
self.assertEqual(actual, expected)
def test_rss_zero(self):
'''
.rss() should be zero due to being identical.
'''
Y1 = np.arange(100)
Y2 = np.arange(100)
actual = lms.rss(Y1, Y2)
expected = 0
self.assertEqual(actual, expected)
Y1 = np.linspace(0, 100, 3)
Y2 = np.linspace(0, 100, 3)
self.assertEqual(actual, expected)
def test_standard_error_advertising_full_X(self):
'''
This is testing standard error for all three columns using data from ISLR pg. 74 to check against
'''
standard_error_data = [('TV', 0.001395), ('radio', 0.008611),
('newspaper', 0.005871)]
places = 4
df = pd.read_csv('Advertising.csv')
# Set Y to sales
Y = df['sales']
# Set X to newspaper
X = pd.DataFrame(df[['TV', 'radio', 'newspaper']])
# Create a linear model and train
reg = LinearRegression()
reg.fit(X, Y)
# Get predictions for Y
Y_pred = reg.predict(X)
rss = lms.rss(Y.values, Y_pred)
rse = lms.rse(rss, Y.values.size, X.columns.size)
var = rse**2
for i in range(len(standard_error_data)):
stan_e_tup = standard_error_data[i]
standard_error = lms.standard_error(var, X[stan_e_tup[0]].values)
actual = standard_error
expected = stan_e_tup[1]
print(stan_e_tup[0])
print('Actual:\t\t', actual)
print('Expected:\t', expected)
print()
def test_rse_basic(self):
'''
This test is testing rse in a basic case can be easily calculated manually.
'''
Y1 = np.array([1, 3, 7, 11])
Y2 = np.array([0, 6, 9, 11])
# rss = 14
rss = lms.rss(Y1, Y2)
actual = lms.rse(rss, Y1.size, 2)
# Should be sqrt of 14
expected = 3.74165738677
self.assertAlmostEqual(actual, expected, places=7)
actual = lms.rse(rss, Y1.size, 1)
# Should be sqrt of 7
expected = 2.64575131106
self.assertAlmostEqual(actual, expected, places=7)