def test_nn(self):
vw = VW(convert_to_vw=False, nn=3)
pos = "1.0 | a b c"
neg = "-1.0 | d e f"
vw.fit([pos] * 10 + [neg] * 10)
assert vw.predict(["| a b c"]) > 0
assert vw.predict(["| d e f"]) < 0
def test_predict(self, data):
raw_model = VW()
raw_model.fit(data.x, data.y)
model = VWRegressor()
model.fit(data.x, data.y)
assert np.allclose(raw_model.predict(data.x), model.predict(data.x))
# ensure model can make multiple calls to predict
assert np.allclose(raw_model.predict(data.x), model.predict(data.x))
def test_passes(self, data):
n_passes = 2
model = VW(loss_function="logistic", passes=n_passes)
assert getattr(model, "passes") == n_passes
model.fit(data.x, data.y)
weights = model.get_coefs()
model = VW(loss_function="logistic")
# first pass weights should not be the same
model.fit(data.x, data.y)
assert not np.allclose(weights.data, model.get_coefs().data)
def test_oaa(self):
X = [
"1 | feature1:2.5",
"2 | feature1:0.11 feature2:-0.0741",
"3 | feature3:2.33 feature4:0.8 feature5:-3.1",
"1 | feature2:-0.028 feature1:4.43",
"2 | feature5:1.532 feature6:-3.2",
]
model = VW(convert_to_vw=False, oaa=3, loss_function="logistic")
model.fit(X)
prediction = model.predict(X)
assert np.allclose(prediction, [1.0, 2.0, 3.0, 1.0, 2.0])
def test_save_load(self, data):
file_name = "tmp_sklearn.model"
model_before = VW(l=100)
model_before.fit(data.x, data.y)
before_saving = model_before.predict(data.x)
model_before.save(file_name)
model_after = VW(l=100)
model_after.load(file_name)
after_loading = model_after.predict(data.x)
assert np.allclose(before_saving, after_loading)
def test_lrq(self):
X = [
"1 |user A |movie 1",
"2 |user B |movie 2",
"3 |user C |movie 3",
"4 |user D |movie 4",
"5 |user E |movie 1",
]
model = VW(convert_to_vw=False,
lrq="um4",
lrqdropout=True,
loss_function="quantile")
assert getattr(model, "lrq") == "um4"
assert getattr(model, "lrqdropout")
model.fit(X)
prediction = model.predict([" |user C |movie 1"])
assert np.allclose(prediction, [3.0], atol=1)
def test_oaa_probs(self):
X = [
"1 | feature1:2.5",
"2 | feature1:0.11 feature2:-0.0741",
"3 | feature3:2.33 feature4:0.8 feature5:-3.1",
"1 | feature2:-0.028 feature1:4.43",
"2 | feature5:1.532 feature6:-3.2",
]
model = VW(convert_to_vw=False,
oaa=3,
loss_function="logistic",
probabilities=True)
model.fit(X)
prediction = model.predict(X)
assert prediction.shape[0] == 5
assert prediction.shape[1] == 3
assert prediction[0, 0] > 0.1
def test_bfgs(self):
data_file = os.path.join(os.path.dirname(os.path.realpath(__file__)),
"resources", "train.dat")
model = VW(
convert_to_vw=False,
oaa=3,
passes=30,
bfgs=True,
data=data_file,
cache=True,
quiet=False,
)
model.fit()
X = [
"1 | feature1:2.5",
"2 | feature1:0.11 feature2:-0.0741",
"3 | feature3:2.33 feature4:0.8 feature5:-3.1",
"1 | feature2:-0.028 feature1:4.43",
"2 | feature5:1.532 feature6:-3.2",
]
actual = model.predict(X)
assert np.allclose(actual, [1.0, 2.0, 3.0, 1.0, 2.0])
def test_fit(self, data):
model = VW(loss_function="logistic")
assert model.vw_ is None
model.fit(data.x, data.y)
assert model.vw_ is not None
def test_del(self, data):
model = VW()
model.fit(data.x, data.y)
del model
def test_get_intercept(self, data):
model = VW()
model.fit(data.x, data.y)
intercept = model.get_intercept()
assert isinstance(intercept, float)
def test_get_coefs(self, data):
model = VW()
model.fit(data.x, data.y)
weights = model.get_coefs()
assert np.allclose(weights.indices,
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 116060])
def test_predict_no_convert(self):
model = VW(loss_function="logistic", convert_to_vw=False)
model.fit(["-1 | bad", "1 | good"])
actual = model.predict(["| good"])[0]
assert np.isclose(actual, 0.245515, atol=1e-2)
def test_predict(self, data):
model = VW(loss_function="logistic")
model.fit(data.x, data.y)
actual = model.predict(data.x[:1][:1])[0]
assert np.isclose(actual, 0.406929, atol=1e-2)