AttributeError: 'numpy.ndarray' object has no attribute 'transform'

Nobody has claimed this yet.

Hello,

When trying to do a `.fit_transform` on the "Train" data and then a `.transform` on the "Test" data, I get the error:

AttributeError: 'numpy.ndarray' object has no attribute 'transform'

This is my use case:

```
reducer = umap.UMAP(
n_components=11, n_neighbors=20, metric = "manhattan", random_state=42, transform_seed=42, verbose=False
).fit_transform(X_train, y_train)

reducer_test = reducer.transform(X_test)
```

Not sure if this is a bug that the `.fit_transform` and `.transform` can't be used for the "Train" and "Test" data respectively.

Can you tell me how to overcome this error? Thanks!

Side note: Not sure if this is related, but I can overcome the error by removing the `_transform` for the "Train" , however, when I use a random forest, I get just an 82% accuracy (even after doing a grid search for the hyper-parameters). Is it a fair assumption that as I do not `transform` the "Test" data, that the accuracy may be lower than expected?

Start at the `umap.UMAP` usage shown in the issue and compare the object returned by `.fit_transform` with the fitted reducer needed for `.transform(X_test)`. Reproduce the train/test example with Python and numpy; done means the issue clarifies the correct fitting and transformation flow and addresses the accuracy question.

Written by the indexing model from the issue text.