[REVIEW] 018 add unfitted error pca & tests on IPCA

python/cuml/decomposition/pca.pyx

@@ -46,6 +46,7 @@ from cuml.common.array_descriptor import CumlArrayDescriptor
 from cuml.common import using_output_type
 from cuml.prims.stats import cov
 from cuml.common.input_utils import sparse_scipy_to_cp
+from cuml.common.exceptions import NotFittedError
 
 
 cdef extern from "cuml/decomposition/pca.hpp" namespace "ML":
@@ -554,6 +555,7 @@ class PCA(Base):
 
         """
 
+        self._check_is_fitted('components_')
         if cupyx.scipy.sparse.issparse(X):
             return self._sparse_inverse_transform(X,
                                                   return_sparse=return_sparse,
@@ -653,6 +655,7 @@ class PCA(Base):
 
         """
 
+        self._check_is_fitted('components_')
         if cupyx.scipy.sparse.issparse(X):
             return self._sparse_transform(X)
         elif scipy.sparse.issparse(X):
@@ -736,3 +739,9 @@ class PCA(Base):
             'X_types_gpu': ['2darray', 'sparse'],
             'X_types': ['2darray', 'sparse']
         }
+
+    def _check_is_fitted(self, attr):
+        if not hasattr(self, attr) or (getattr(self, attr) is None):
+            msg = ("This instance is not fitted yet. Call 'fit' "
+                   "with appropriate arguments before using this estimator.")
+            raise NotFittedError(msg)

python/cuml/experimental/decomposition/incremental_pca.py

@@ -627,7 +627,7 @@ def _svd_flip(u, v, u_based_decision=True):
     if u_based_decision:
         # columns of u, rows of v
         max_abs_cols = cp.argmax(cp.abs(u), axis=0)
-        signs = cp.sign(u[max_abs_cols, range(u.shape[1])])
+        signs = cp.sign(u[max_abs_cols, list(range(u.shape[1]))])
         u *= signs
         v *= signs[:, cp.newaxis]
     else:

python/cuml/test/test_incremental_pca.py

@@ -22,8 +22,10 @@
 
 from cuml.datasets import make_blobs
 from cuml.experimental.decomposition import IncrementalPCA as cuIPCA
+from cuml.experimental.decomposition.incremental_pca import _svd_flip
 
 from cuml.test.utils import array_equal
+from cuml.common.exceptions import NotFittedError
 
 
 @pytest.mark.parametrize(
@@ -109,3 +111,35 @@ def test_partial_fit(nrows, ncols, n_components, density,
 
     assert array_equal(cu_inv, sk_inv,
                        5e-5, with_sign=True)
+
+
+def test_exceptions():
+    X = cupyx.scipy.sparse.eye(10)
+    ipca = cuIPCA()
+    with pytest.raises(TypeError):
+        ipca.partial_fit(X)
+
+    X = X.toarray()
+    with pytest.raises(NotFittedError):
+        ipca.transform(X)
+
+    with pytest.raises(NotFittedError):
+        ipca.inverse_transform(X)
+
+    with pytest.raises(ValueError):
+        cuIPCA(n_components=8).fit(X[:5])
+
+    with pytest.raises(ValueError):
+        cuIPCA(n_components=8).fit(X[:, :5])
+
+
+def test_svd_flip():
+    x = cp.array(range(-10, 80)).reshape((9, 10))
+    u, s, v = cp.linalg.svd(x, full_matrices=False)
+    u_true, v_true = _svd_flip(u, v, u_based_decision=True)
+    reco_true = cp.dot(u_true * s, v_true)
+    u_false, v_false = _svd_flip(u, v, u_based_decision=False)
+    reco_false = cp.dot(u_false * s, v_false)
+
+    assert array_equal(reco_true, x)
+    assert array_equal(reco_false, x)

python/cuml/test/test_pca.py

@@ -26,6 +26,7 @@
 from sklearn.datasets import make_multilabel_classification
 from sklearn.decomposition import PCA as skPCA
 from sklearn.datasets import make_blobs
+from cuml.common.exceptions import NotFittedError
 
 
 @pytest.mark.parametrize('datatype', [np.float32, np.float64])
@@ -233,3 +234,13 @@ def test_sparse_pca_inputs(nrows, ncols, whiten, return_sparse, cupy_input):
             assert isinstance(i_sparse, cp.core.ndarray)
 
         assert array_equal(i_sparse, X.todense(), 1e-1, with_sign=True)
+
+
+def test_exceptions():
+    with pytest.raises(NotFittedError):
+        X = cp.random.random((10, 10))
+        cuPCA().transform(X)
+
+    with pytest.raises(NotFittedError):
+        X = cp.random.random((10, 10))
+        cuPCA().inverse_transform(X)