Feature request
PHATE.transform(X) raises a ValueError when the model is initialized with knn_dist="precomputed_affinity".
This configuration should be supported natively, as the PHATE out-of-sample projection only requires an affinity matrix between query samples and training samples (or training landmarks). Therefore, precomputed_affinity is a valid input for PHATE.transform() and should not trigger an error.
To Reproduce
import forestgeom
import phate
from forestgeom import ForestProximity
from sklearn.datasets import load_iris
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
iris = load_iris()
x = iris.data
y = iris.target
x_train, x_test, y_train, y_test = train_test_split(
x, y, test_size=0.25, random_state=42, stratify=y
)
forest = RandomForestClassifier(random_state=42, n_estimators=100)
proximity_model = ForestProximity(forest=forest, weight_scheme="gap")
proximity_model.fit(x_train, y_train)
train_affinities = proximity_model.training_proximity(adjust_diagonal=True) # sparse NxN training affinities
phate_op = phate.PHATE(
knn_dist="precomputed_affinity",
n_components=2,
n_landmark=10,
random_state=42,
verbose=1,
)
emb_train = phate_op.fit_transform(train_affinities)
test_train_affinities = proximity_model.transform(x_test) # sparse N_test x N affinity matrix
emb_test = phate_op.transform(test_train_affinities)
Expected behavior
For any input affinity matrix X of shape (n_query, n_train) and a PHATE operator initialized with knn_dist="precomputed_affinity", PHATE.transform(X) should perform the standard PHATE out-of-sample extension.
Case 1: Full Diffusion Operator
If PHATE was fitted without landmarks, the output embedding should be
Z_query = P_query-train · Z_train
where:
P_query-train has shape (n_query, n_train),P_query-train is obtained by row-normalizing the input affinity matrix X,Z_train is the PHATE embedding of the training samples.
Case 2: Landmark Diffusion Operator
If PHATE was fitted using landmarks, the output embedding should be
Z_query = P_query-landmark · Z_landmark
where:
P_query-landmark has shape (n_query, n_landmark),P_query-landmark is obtained by aggregating the affinities in X according to the landmark assignments computed during landmark operator construction, followed by row normalization,Z_landmark is the PHATE embedding of the landmark points.
Actual behavior
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In[3], line 32
28 )
29 emb_train = phate_op.fit_transform(train_affinities)
30
31 test_train_affinities = proximity_model.transform(x_test) # sparse N_test x N affinity matrix
---> 32 emb_test = phate_op.transform(test_train_affinities)
File ~/Projects/RF-PHATE/.venv/lib/python3.14/site-packages/phate/phate.py:980, in PHATE.transform(self, X, t_max, plot_optimal_t, ax)
970 warnings.warn(
971 "Pre-fit PHATE should not be used to transform a "
972 "new data matrix. Please fit PHATE to the new"
973 " data by running 'fit' with the new data.",
974 RuntimeWarning,
975 )
976 if (
977 isinstance(self.graph, graphtools.graphs.TraditionalGraph)
978 and self.graph.precomputed is not None
979 ):
--> 980 raise ValueError(
981 "Cannot transform additional data using a "
982 "precomputed distance matrix."
983 )
984 else:
985 if self.embedding is None:
ValueError: Cannot transform additional data using a precomputed distance matrix.
System information:
Output of phate.__version__:
Output of forestgeom.__version__:
Output of pd.show_versions():
Details
INSTALLED VERSIONS
------------------
commit : 72f2fea91530b5abb3cf2100cb22d84e504695c0
python : 3.14.5
python-bits : 64
OS : Darwin
OS-release : 25.5.0
Version : Darwin Kernel Version 25.5.0: Mon Apr 27 20:41:15 PDT 2026; root:xnu-12377.121.6~2/RELEASE_ARM64_T6041
machine : arm64
processor : arm
byteorder : little
LC_ALL : None
LANG : C.UTF-8
LOCALE : C.UTF-8
pandas : 3.0.3
numpy : 2.5.0
dateutil : 2.9.0.post0
pip : None
Cython : None
sphinx : None
IPython : 9.14.0
adbc-driver-postgresql: None
adbc-driver-sqlite : None
bs4 : None
bottleneck : None
fastparquet : None
fsspec : None
html5lib : None
hypothesis : None
gcsfs : None
jinja2 : None
lxml.etree : None
matplotlib : 3.10.9
numba : None
numexpr : None
odfpy : None
openpyxl : None
psycopg2 : None
pymysql : None
pyarrow : None
pyiceberg : None
pyreadstat : None
pytest : None
python-calamine : None
pytz : None
pyxlsb : None
s3fs : None
scipy : 1.18.0
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlsxwriter : None
zstandard : None
qtpy : None
pyqt5 : None
Feature request
PHATE.transform(X)raises aValueErrorwhen the model is initialized withknn_dist="precomputed_affinity".This configuration should be supported natively, as the PHATE out-of-sample projection only requires an affinity matrix between query samples and training samples (or training landmarks). Therefore,
precomputed_affinityis a valid input forPHATE.transform()and should not trigger an error.To Reproduce
Expected behavior
For any input affinity matrix
Xof shape(n_query, n_train)and a PHATE operator initialized withknn_dist="precomputed_affinity",PHATE.transform(X)should perform the standard PHATE out-of-sample extension.Case 1: Full Diffusion Operator
If PHATE was fitted without landmarks, the output embedding should be
where:
P_query-trainhas shape(n_query, n_train),P_query-trainis obtained by row-normalizing the input affinity matrixX,Z_trainis the PHATE embedding of the training samples.Case 2: Landmark Diffusion Operator
If PHATE was fitted using landmarks, the output embedding should be
where:
P_query-landmarkhas shape(n_query, n_landmark),P_query-landmarkis obtained by aggregating the affinities inXaccording to the landmark assignments computed during landmark operator construction, followed by row normalization,Z_landmarkis the PHATE embedding of the landmark points.Actual behavior
System information:
Output of
phate.__version__:Output of
forestgeom.__version__:Output of
pd.show_versions():Details