def _learn_granular(self, feat_mat, grades):
min_grade = min(grades)
max_grade = max(grades)
grade_to_model = {}
cur_center_grade = min_grade + 1
while cur_center_grade < max_grade:
inds_within_one_grade = [ind for ind, grade in enumerate(grades) if abs(grade - cur_center_grade) < 1.1]
cur_feat_mat = np.vstack([feat_mat[ind, :] for ind in inds_within_one_grade])
cur_grades = [grades[ind] for ind in inds_within_one_grade]
learner = LinearRegression(intercept=True, debug=params.DEBUG)
learner.train(
cur_feat_mat,
cur_grades,
self.ds_train.getEssaySet(),
self.ds_train.getDomain(),
{"feature_selection": "inclusive"},
)
grade_to_model[cur_center_grade] = learner
cur_center_grade += 1
grade_to_model[min_grade] = grade_to_model[min_grade + 1]
grade_to_model[max_grade] = grade_to_model[max_grade - 1]
return grade_to_model
def test_correlations(self):
X = np.array([[1, 1], [2, 3], [3, 1]])
Y = np.array([1, 3, 1])
lr = LinearRegression()
correlations = lr.get_feature_grade_correlations(X, Y)
expected = [0, 1]
for feat_ind, corr in enumerate(correlations):
self.assertAlmostEqual(expected[feat_ind], corr)
def test_predict(self):
X = [[1, 0], [0, 2]]
Y = [3, 2]
lr = LinearRegression()
lr.train(X, Y)
self.assertAlmostEqual(3, lr.predict([1,0]))
self.assertAlmostEqual(4, lr.predict([1,1]))
def test_solve(self):
X = [[1, 0], [0, 2]]
Y = [3, 2]
lr = LinearRegression()
lr.train(X, Y)
self.assertFalse(lr.has_intercept)
self.assertAlmostEqual(3, lr.params[0])
self.assertAlmostEqual(1, lr.params[1])
self.assertEqual(2, len(lr.params))
X = [[1, 0], [0, 2], [0, 3]]
Y = [3, 2, 1]
lr = LinearRegression(intercept=True)
lr.train(X, Y)
self.assertTrue(lr.has_intercept)
self.assertAlmostEqual(-1, lr.params[0])
self.assertAlmostEqual(-1, lr.params[1])
self.assertEqual(2, len(lr.params))
self.assertAlmostEqual(4, lr.intercept)