def test_pick_custom_emb_dim(test_data):
cat_sz = categorize(test_data)
emb_sz = pick_emb_dim(cat_sz, emb_dims=[3, 4])
assert len(emb_sz) == 2
assert emb_sz['Categorical 1'] == (10, 3)
assert emb_sz['Categorical 2'] == (10, 4)
def test_categorize(test_data):
cat_sz = categorize(test_data)
assert len(cat_sz) == 2
assert cat_sz[0] == ('Categorical 1', 10)
assert cat_sz[1] == ('Categorical 2', 10)
def test_pick_emb_dim(test_data):
cat_sz = categorize(test_data)
emb_sz = pick_emb_dim(cat_sz)
assert len(emb_sz) == 2
assert emb_sz['Categorical 1'] == (10, 5)
assert emb_sz['Categorical 2'] == (10, 5)
def test_transform(test_data):
X, y = test_data
cat_sz = categorize(X)
emb_sz = pick_emb_dim(cat_sz)
X, encoders = encode_categorical(X)
embedder = Embedder(emb_sz)
embedder.fit(X, y, epochs=1)
transformed = embedder.transform(X)
assert transformed.shape == (10000, 18 + 50 + 50)
def test_fit_predict(test_data):
X, y = test_data
cat_sz = categorize(X)
emb_sz = pick_emb_dim(cat_sz)
X, encoders = encode_categorical(X)
embedder = Embedder(emb_sz)
embedder.fit(X, y, epochs=1)
preds = embedder.predict(X[:100])
assert len(preds) == 100
assert all(preds <= 1) and all(preds >= 0)
def test_fit_predict(test_data):
X, y = test_data
cat_sz = categorize(X)
emb_sz = pick_emb_dim(cat_sz)
X, encoders = encode_categorical(X)
embedder = Embedder(emb_sz)
embedder.fit(X, y, epochs=1)
preds = embedder.predict(X[:100])
assert len(preds) == 100
assert not np.isinf(preds).any()
assert not np.isnan(preds).any()
def test_pipeline(test_data):
X, y = test_data
cat_sz = categorize(X)
emb_sz = pick_emb_dim(cat_sz)
X_encoded, encoders = encode_categorical(X)
pipeline = Pipeline([('embedding', Embedder(emb_sz)),
('randomforest', RandomForestRegressor())])
pipeline.fit(X_encoded, y)
preds = pipeline.predict(X_encoded)
assert len(preds) == 10000
assert not np.isinf(preds).any()
assert not np.isnan(preds).any()