def test_computing_euclidean_distance_with_missing_values():
dummy_train = np.array([[0.2, 0.5, 1.0], [-1, 0.4, 0.1], [0.2, 0.4, 0.4]])
dummy_test = np.array([[0.8, 0.1, -1], [0.7, 0.2, 0.4], [0.6, -1, 0.8]])
distance = euclidean_distances(dummy_test, dummy_train)
print(distance)
#assert_array_equal(expected_distance, distance)
missing_indexes = KNNAlgorithmM().get_missing_values_indexes(dummy_test)
print(missing_indexes)
missing_train_indexes = KNNAlgorithmM().get_missing_values_indexes(
dummy_train)
print(missing_train_indexes)
for k in missing_indexes.keys():
for test in range(dummy_train.shape[0]):
distance[int(k)][test] = KNNAlgorithmM(
).euclidean_distance_with_missing_values(dummy_test[test],
dummy_train[int(k)])
print(distance)
#for k in missing_train_indexes.keys():
# for train in range(dummy_test.shape[0]):
# distance[train][int(k)] = KNNAlgorithmM.euclidean_distance_with_missing_values(dummy_train[int(k)], dummy_test[train])
#print(distance)
i = j = 0
for test in dummy_test:
for train in dummy_train:
distance[i][j] = KNNAlgorithmM(
).euclidean_distance_with_missing_values(test, train)
j += 1
i += 1
j = 0
print(distance)
expected_distance = np.array([[1.232, 1.240, 0.9], [0.836, 0.787, 0.538],
[0.67, 1.191, 0.824]])
assert_array_equal(expected_distance, distance)
def test_classifier(data):
# tests default attributes
X, y = data
aggregatedKNN = KNNAlgorithmM()
assert aggregatedKNN.k_neighbours == (3, 5, 7)
# tests adding special fields in learning
aggregatedKNN.fit(X, y)
assert hasattr(aggregatedKNN, 'classes_')
assert hasattr(aggregatedKNN, 'X_')
assert hasattr(aggregatedKNN, 'y_')
# tests output shape
y_pred = aggregatedKNN.predict(X)
assert y_pred.shape == (X.shape[0], )
def test_main_class_probability():
test_decisions = np.array([0, 0, 1, 1, 0, 1, 0, 1, 1, 0, 0, 1])
expected_p = np.array([[1 / 3, 2 / 5, 3 / 7], [1 / 3, 3 / 5, 4 / 7],
[1, 3 / 5, 5 / 7]])
a = KNNAlgorithmM()
a.y_ = test_decisions
sorted_distance = np.array([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11],
[5, 6, 1, 2, 3, 0, 7, 9, 8, 10, 11, 4],
[3, 7, 8, 9, 10, 11, 2, 1, 4, 0, 6, 5]])
p = np.zeros(shape=(3, 3))
j = 0
for i in range(0, 3):
j = 0
for k in (3, 5, 7):
p[i, j] = a._classic_knn(sorted_distance, k, i)
j += 1
assert_allclose(p, expected_p, rtol=0.1, atol=0.1)
def test_measure():
# create in-memory dataset
bmi = pd.DataFrame.from_dict({
'height': [1.6, 1.6, 1.62, 1.75, 1.7, 1.8, 1.9],
'weight': [60, 80, 80, 90, 80, 85, 82],
'bmi': [1, 0, 0, 0, 0, 1, 1]
})
print(bmi)
aggregatedKNN = KNNAlgorithmM(k_neighbours=(1, 3))
# fit a model
model = aggregatedKNN.fit(bmi.iloc[:, :2], np.ravel(bmi.iloc[:, 2:3]))
print(bmi.iloc[:, :2])
print(np.ravel(bmi.iloc[:, 2:3]))
# checking first object decision
pred = model.predict([[1.6, 60]])
print(pred)
#assert pred == [1]
# checking classification score (accuracy here)
score = model.score([[1.6, 60], [1.59, 61], [1.6, 80]], [1, 1, 0])
print(score)
assert score == 1.0
def test_m_dist_equivalence():
X, y = load_iris(True)
m = KNNAlgorithmM()
m.fit(X, y)
def full_distance(X, X_):
distance = np.empty(shape=(X.shape[0], X_.shape[0]), dtype=np.float64)
i = j = 0
for test in X:
for train in X_:
distance[i][j] = m.euclidean_distance_with_missing_values(
test, train)
j += 1
i += 1
j = 0
return distance
#assert_array_equal(m.euclidean_distance_with_missing_values(X, X),
# m.euclidean_distance_with_missing_values_optimized(X, X))
assert_array_equal(
m.euclidean_distance_with_missing_values_optimized(X, X),
full_distance(X, X))
.fit_transform(X)
multiclassF = OneVsRestClassifierForRandomBinaryClassifier(KNNAlgorithmF(missing_representation=-1, r=r, aggregation=agg, k_neighbours=k))
f_result = cross_validate(multiclassF, X_missing, y, scoring='roc_auc_ovo', return_estimator=True, cv=10)
w = pd.DataFrame({'algorithm': 'f', 'k': str(k), 'r': f_result['estimator'][0].estimator.r,
'agg': Aggregation.change_aggregation_to_name(f_result['estimator'][0].estimator.aggregation), 'missing': miss,
'auc': np.mean(f_result['test_score']), 'stddev': stdev(f_result['test_score'])}, index=[ind])
print(w)
dfs.append(w)
ind += 1
concatenated = pd.concat(dfs)
concatenated = pd.concat(dfs)
concatenated.to_excel('concatenated_vertebral.xlsx')
for miss in missing:
for k in ks:
X_missing = MissingValuesInserterColumnsIndependent(columns=range(X.shape[1]), nan_representation=-1, percentage=miss) \
.fit_transform(X)
multiclassM = OneVsRestClassifier(KNNAlgorithmM(missing_representation=-1, k_neighbours=k))
m_result = cross_validate(multiclassM, X_missing, y, scoring='roc_auc_ovo', return_estimator=True, cv=10)
w2 = pd.DataFrame({'algorithm': 'm', 'k': str(k), 'r': '',
'agg': '',
'missing': miss,
'auc': np.mean(m_result['test_score']), 'stddev': stdev(m_result['test_score'])},
index=[ind])
print(w2)
dfs.append(w2)
ind += 1
concatenated = pd.concat(dfs)
concatenated.to_excel('concatenated_vertebral_m.xlsx')
stdev(f_result['test_score'])
},
index=[ind])
print(w)
dfs.append(w)
ind += 1
concatenated = pd.concat(dfs)
concatenated.to_excel('concatenated_iris.xlsx')
for miss in missing:
for k in ks:
X_missing = MissingValuesInserterColumnsIndependent(columns=range(X.shape[1]), nan_representation=-1, percentage=miss) \
.fit_transform(X)
multiclassM = OneVsRestClassifier(
KNNAlgorithmM(missing_representation=-1, k_neighbours=k))
m_result = cross_validate(multiclassM,
X_missing,
y,
scoring='roc_auc_ovo',
return_estimator=True,
cv=10)
w2 = pd.DataFrame(
{
'algorithm': 'm',
'k': str(k),
'r': '',
'agg': '',
'missing': miss,
'auc': np.mean(m_result['test_score']),
def test_sorting():
distance = np.array([[0.8653, 0.6731, 0.9863, 0.3042, 0.9, 0.452],
[0.1231, 0.8763, 0.0112, 0.7633, 0.912, 0.341]])
tested = KNNAlgorithmM().take_k_smallest_and_sort_distances(distance, 5)
assert_array_equal(tested, np.argsort(distance)[:, :5])