The parameter init="pca" cannot be used with metric="precomputed".

ValueError: The parameter init="pca" cannot be used with metric="precomputed".

Changed in version 1.2: The default value changed to "pca".

model = TSNE(n_components=2, random_state=0, metric='precomputed',
                 perplexity=min(len(dist_matrix) - 1, 30), **kwargs)

add back what was the default param initially: init='random'

See also scikit-learn/scikit-learn#18018

Author: msusol

HISTORY.rst

@@ -9,6 +9,7 @@ History
 * Adding testing for Python 3.10, 3.11, move default version to Python 3.10.
 * Tox testing: No module named 'sklearn.manifold'; 'sklearn' is not a package.
   * Rename sklearn.py to lda_model.py.
+* ValueError: The parameter init="pca" cannot be used with metric="precomputed".
 * Update sklearn.py #239.
 * fixes error of get_feature_names removal #235.
 * Remove "sklearn" from requirements #234

notebooks/LDA model.ipynb

@@ -158,7 +158,7 @@ def js_TSNE(distributions, **kwargs):
     tsne : array, shape (`n_dists`, 2)
     """
     dist_matrix = squareform(pdist(distributions, metric=_jensen_shannon))
-    model = TSNE(n_components=2, random_state=0, metric='precomputed',
+    model = TSNE(n_components=2, random_state=0, metric='precomputed', init='random',
                  perplexity=min(len(dist_matrix) - 1, 30), **kwargs)
     return model.fit_transform(dist_matrix)