Shift of UMAP points when using .transform()

Nobody has claimed this yet.

Hi,
I am encountering a point shift issue when fitting umap to embeddings X_train and then applying the fit to embeddings X_test not seen during training.

Here is an example data set creation to demonstrate the issue:

The new points are all shifted to the bottom left corner of the UMAP space:

This is the code for the example above
```import numpy as np
import pandas as pd
import plotly.express as px
from umap import UMAP
n_examples = 1000
X = np.random.random((n_examples,128))
X_train = X[:n_examples//2]
X_test = X[n_examples//2:]
mapper = UMAP()
mapper.fit(X_train)
low_dim_emb_train = mapper.transform(X_train)
low_dim_emb_test = mapper.transform(X_test)
df_train = pd.DataFrame({"x":low_dim_emb_train[:,0],"y":low_dim_emb_train[:,1],"dataset":["train"]*len(low_dim_emb_train)})
df_test = pd.DataFrame({"x":low_dim_emb_test[:,0],"y":low_dim_emb_test[:,1],"dataset":["test"]*len(low_dim_emb_test)})
df_plot = pd.concat([df_train,df_test])
px.scatter(df_plot,x="x",y="y",color="dataset")
```

What could be the issue?
Many thanks
Olivia

Start by running the supplied Python reproducer with UMAP.fit() on X_train and transform() on both X_train and X_test, then inspect the transform behavior and related UMAP implementation entry points. Done means explaining or correcting the reported bottom-left shift for unseen points and adding a regression check for the demonstrated case; no specific file or test is named in the issue.

Written by the indexing model from the issue text.