def test_digits_corr_lazy_init(): model = FacilityLocationSelection(100, 'corr', optimizer='lazy', initial_subset=digits_corr_ranking[:5]) model.fit(X_digits) assert_array_equal(model.ranking[:-5], digits_corr_ranking[5:]) assert_array_almost_equal(model.gains[:-5], digits_corr_gains[5:], 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_cosine_two_stage_init(): model = FacilityLocationSelection(100, 'cosine', optimizer='two-stage', initial_subset=digits_cosine_ranking[:5]) model.fit(X_digits) assert_array_equal(model.ranking[:-5], digits_cosine_ranking[5:]) assert_array_almost_equal(model.gains[:-5], digits_cosine_gains[5:], 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_euclidean_naive_init(): model = FacilityLocationSelection(100, 'euclidean', optimizer='naive', initial_subset=digits_euclidean_ranking[:5]) model.fit(X_digits) assert_array_equal(model.ranking[:20], digits_euclidean_ranking[5:25]) assert_array_almost_equal(model.gains[:20], digits_euclidean_gains[5:25], 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_sqrt_modular_sparse(): model = FacilityLocationSelection(100, 'precomputed', optimizer='modular', random_state=0) model.fit(X_digits_cosine_sparse) assert_array_equal(model.ranking, digits_cosine_modular_ranking) assert_array_almost_equal(model.gains, digits_cosine_modular_gains, 4)
def test_digits_corr_small_pq_bool_initialized():
mask = numpy.zeros(X_digits.shape[0], dtype=bool)
mask[digits_corr_ranking[:5]] = True
model = FacilityLocationSelection(10, 'corr', 1, initial_subset=mask)
model.fit(X_digits)
assert_array_equal(model.ranking, digits_corr_ranking[5:15])
assert_array_almost_equal(model.gains, digits_corr_gains[5:15], 4)
def test_digits_cosine_approximate_object():
model = FacilityLocationSelection(100,
'cosine',
optimizer=ApproximateLazyGreedy())
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_approx_ranking)
assert_array_almost_equal(model.gains, digits_cosine_approx_gains, 4)
def test_digits_cosine_approximate_sparse():
model = FacilityLocationSelection(100,
'precomputed',
optimizer='approximate-lazy')
model.fit(X_digits_cosine_sparse)
assert_array_equal(model.ranking, digits_cosine_approx_ranking)
assert_array_almost_equal(model.gains, digits_cosine_approx_gains, 4)
def test_digits_sqrt_modular_object():
model = FacilityLocationSelection(100,
'cosine',
optimizer=ModularGreedy(random_state=0))
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_modular_ranking)
assert_array_almost_equal(model.gains, digits_cosine_modular_gains, 4)
def test_digits_cosine_two_stage_sparse():
model = FacilityLocationSelection(100,
'precomputed',
optimizer='two-stage')
model.fit(X_digits_cosine_sparse)
assert_array_equal(model.ranking, digits_cosine_ranking)
assert_array_almost_equal(model.gains, digits_cosine_gains, 4)
def test_digits_cosine_greedi_nl_sparse():
model = FacilityLocationSelection(100, 'precomputed', optimizer='greedi',
optimizer_kwds={'optimizer1': 'naive', 'optimizer2': 'lazy'},
random_state=0)
model.fit(X_digits_cosine_sparse)
assert_array_equal(model.ranking[:30], digits_cosine_greedi_ranking[:30])
assert_array_almost_equal(model.gains[:30], digits_cosine_greedi_gains[:30], 4)
def test_digits_euclidean_naive(): model = FacilityLocationSelection(100, 'euclidean', optimizer='naive') model.fit(X_digits) assert_array_equal(model.ranking[:30], digits_euclidean_ranking[:30]) assert_array_equal(model.ranking[-30:], digits_euclidean_ranking[-30:]) assert_array_almost_equal(model.gains, digits_euclidean_gains, 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_cosine_two_stage_object():
model = FacilityLocationSelection(100,
'cosine',
optimizer=TwoStageGreedy())
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_ranking)
assert_array_almost_equal(model.gains, digits_cosine_gains, 4)
def test_digits_cosine_sample_object(): model = FacilityLocationSelection(100, 'cosine', optimizer=SampleGreedy(random_state=0)) model.fit(X_digits) assert_array_equal(model.ranking, digits_cosine_sample_ranking) assert_array_almost_equal(model.gains, digits_cosine_sample_gains, 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_cosine_greedi_nl_object(): model = FacilityLocationSelection(100, 'cosine', optimizer=GreeDi( optimizer1='naive', optimizer2='lazy', random_state=0)) model.fit(X_digits) assert_array_equal(model.ranking[:30], digits_cosine_greedi_ranking[:30]) assert_array_almost_equal(model.gains[:30], digits_cosine_greedi_gains[:30], 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_cosine_modular(): model = FacilityLocationSelection(100, 'cosine', optimizer='modular', random_state=0) model.fit(X_digits) assert_array_equal(model.ranking, digits_cosine_modular_ranking) assert_array_almost_equal(model.gains, digits_cosine_modular_gains, 4) assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_corr_small_pq_rank_initialized():
model = FacilityLocationSelection(10,
'corr',
1,
initial_subset=digits_corr_ranking[:5])
model.fit(X_digits)
assert_array_equal(model.ranking, digits_corr_ranking[5:15])
assert_array_almost_equal(model.gains, digits_corr_gains[5:15], 4)
def test_digits_cosine_sample():
model = FacilityLocationSelection(100,
'cosine',
optimizer='sample',
random_state=0)
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_sample_ranking)
assert_array_almost_equal(model.gains, digits_cosine_sample_gains, 4)
def test_digits_precomputed_lazy_init():
model = FacilityLocationSelection(100,
'precomputed',
optimizer='lazy',
initial_subset=digits_cosine_ranking[:5])
model.fit(X_digits_cosine_cupy)
assert_array_equal(model.ranking[:-5], digits_cosine_ranking[5:])
assert_array_almost_equal(model.gains[:-5], digits_cosine_gains[5:], 4)
def test_digits_cosine_greedi_ln():
model = FacilityLocationSelection(100, 'cosine', optimizer='greedi',
optimizer_kwds={'optimizer1': 'lazy', 'optimizer2': 'naive'},
random_state=0)
model.fit(X_digits)
assert_array_equal(model.ranking[:50], digits_cosine_greedi_ranking[:50])
assert_array_almost_equal(model.gains[:50], digits_cosine_greedi_gains[:50], 4)
assert_array_almost_equal(model.subset, X_digits[model.ranking])
def test_digits_cosine_greedi_nn_object():
model = FacilityLocationSelection(100,
'cosine',
optimizer=GreeDi(optimizer1='naive',
optimizer2='naive',
random_state=0))
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_greedi_ranking)
assert_array_almost_equal(model.gains, digits_cosine_greedi_gains, 4)
def test_digits_euclidean_large_greedy():
model = FacilityLocationSelection(100, 'euclidean', 100)
model.fit(X_digits)
assert_array_almost_equal(model.gains, digits_euclidean_gains, 4)
# There is one swap in the resulting ranking because two points have
# the same distance. Thus, the gains are the same but this swap means
# that the ranking aren't the same.
assert_array_equal(model.ranking[:25], digits_euclidean_ranking[:25])
assert_array_equal(model.ranking[-25:], digits_euclidean_ranking[-25:])
def f_analysis():
num_subsets = 10
obj = FacilityLocationSelection(num_subsets,
metric='euclidean',
optimizer='lazy')
obj = obj.fit(data)
subsets = obj.transform(data)
def test_digits_cosine_large_precomputed_greedy():
model = FacilityLocationSelection(100, 'precomputed', 100)
model.fit(X_digits_cosine_cupy)
assert_array_equal(model.ranking, digits_cosine_ranking)
assert_array_almost_equal(model.gains, digits_cosine_gains, 4)
def test_digits_cosine_large_precomputed_truncated_pivot():
model = FacilityLocationSelection(150, 'precomputed', 50)
model.fit(X_digits_sparse)
assert_array_equal(model.ranking[:100], digits_cosine_ranking)
assert_array_almost_equal(model.gains[:100], digits_cosine_gains, 4)
def test_digits_corr_small_precomputed_greedy():
model = FacilityLocationSelection(10, 'precomputed', 10)
model.fit(X_digits_corr_cupy)
assert_array_equal(model.ranking, digits_corr_ranking[:10])
assert_array_almost_equal(model.gains, digits_corr_gains[:10], 4)
def test_digits_cosine_small_precomputed_truncated():
model = FacilityLocationSelection(15, 'precomputed', 1)
model.fit(X_digits_sparse)
assert_array_equal(model.ranking[:10], digits_cosine_ranking[:10])
assert_array_almost_equal(model.gains[:10], digits_cosine_gains[:10], 4)
def test_digits_cosine_large_truncated():
model = FacilityLocationSelection(150, 'cosine', 1)
model.fit(X_digits)
assert_array_equal(model.ranking[:100], digits_cosine_ranking)
assert_array_almost_equal(model.gains[:100], digits_cosine_gains, 4)
def test_digits_cosine_large_pivot():
model = FacilityLocationSelection(100, 'cosine', 50)
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_ranking)
assert_array_almost_equal(model.gains, digits_cosine_gains, 4)
def test_digits_cosine_small_truncated_pivot():
model = FacilityLocationSelection(15, 'cosine', 5)
model.fit(X_digits)
assert_array_equal(model.ranking[:10], digits_cosine_ranking[:10])
assert_array_almost_equal(model.gains[:10], digits_cosine_gains[:10], 4)
def test_digits_cosine_small_pq():
model = FacilityLocationSelection(10, 'cosine', 1)
model.fit(X_digits)
assert_array_equal(model.ranking, digits_cosine_ranking[:10])
assert_array_almost_equal(model.gains, digits_cosine_gains[:10], 4)
