scikit-learn
diff --git a/‎examples/miscellaneous/plot_johnson_lindenstrauss_bound.py
Lines changed: 21 additions & 23 deletions b/‎examples/miscellaneous/plot_johnson_lindenstrauss_bound.py
Lines changed: 21 additions & 23 deletions
@@ -33,9 +33,9 @@
 # .. math::
 #    (1 - eps) \|u - v\|^2 < \|p(u) - p(v)\|^2 < (1 + eps) \|u - v\|^2
 #
-# Where u and v are any rows taken from a dataset of shape (n_samples,
-# n_features) and p is a projection by a random Gaussian N(0, 1) matrix
-# of shape (n_components, n_features) (or a sparse Achlioptas matrix).
+# Where `u` and `v` are any rows taken from a dataset of shape `(n_samples,
+# n_features)` and `p` is a projection by a random Gaussian `N(0, 1)` matrix
+# of shape `(n_components, n_features)` (or a sparse Achlioptas matrix).
 #
 # The minimum number of components to guarantees the eps-embedding is
 # given by:
@@ -60,7 +60,7 @@
     min_n_components = johnson_lindenstrauss_min_dim(n_samples_range, eps=eps)
     plt.loglog(n_samples_range, min_n_components, color=color)
 
-plt.legend(["eps = %0.1f" % eps for eps in eps_range], loc="lower right")
+plt.legend([f"eps = {eps:0.1f}" for eps in eps_range], loc="lower right")
 plt.xlabel("Number of observations to eps-embed")
 plt.ylabel("Minimum number of dimensions")
 plt.title("Johnson-Lindenstrauss bounds:\nn_samples vs n_components")
@@ -84,7 +84,7 @@
     min_n_components = johnson_lindenstrauss_min_dim(n_samples, eps=eps_range)
     plt.semilogy(eps_range, min_n_components, color=color)
 
-plt.legend(["n_samples = %d" % n for n in n_samples_range], loc="upper right")
+plt.legend([f"n_samples = {n}" for n in n_samples_range], loc="upper right")
 plt.xlabel("Distortion eps")
 plt.ylabel("Minimum number of dimensions")
 plt.title("Johnson-Lindenstrauss bounds:\nn_components vs eps")
@@ -97,12 +97,12 @@
 # We validate the above bounds on the 20 newsgroups text document
 # (TF-IDF word frequencies) dataset or on the digits dataset:
 #
-# - for the 20 newsgroups dataset some 500 documents with 100k
+# - for the 20 newsgroups dataset some 300 documents with 100k
 #   features in total are projected using a sparse random matrix to smaller
 #   euclidean spaces with various values for the target number of dimensions
 #   ``n_components``.
 #
-# - for the digits dataset, some 8x8 gray level pixels data for 500
+# - for the digits dataset, some 8x8 gray level pixels data for 300
 #   handwritten digits pictures are randomly projected to spaces for various
 #   larger number of dimensions ``n_components``.
 #
@@ -111,25 +111,25 @@
 # this script.
 
 if "--use-digits-dataset" in sys.argv:
-    data = load_digits().data[:500]
+    data = load_digits().data[:300]
 else:
-    data = fetch_20newsgroups_vectorized().data[:500]
+    data = fetch_20newsgroups_vectorized().data[:300]
 
 # %%
 # For each value of ``n_components``, we plot:
 #
 # - 2D distribution of sample pairs with pairwise distances in original
-#   and projected spaces as x and y axis respectively.
+#   and projected spaces as x- and y-axis respectively.
 #
 # - 1D histogram of the ratio of those distances (projected / original).
 
 n_samples, n_features = data.shape
 print(
-    "Embedding %d samples with dim %d using various random projections"
-    % (n_samples, n_features)
+    f"Embedding {n_samples} samples with dim {n_features} using various "
+    "random projections"
 )
 
-n_components_range = np.array([300, 1000, 10000])
+n_components_range = np.array([300, 1_000, 10_000])
 dists = euclidean_distances(data, squared=True).ravel()
 
 # select only non-identical samples pairs
@@ -141,13 +141,13 @@
     rp = SparseRandomProjection(n_components=n_components)
     projected_data = rp.fit_transform(data)
     print(
-        "Projected %d samples from %d to %d in %0.3fs"
-        % (n_samples, n_features, n_components, time() - t0)
+        f"Projected {n_samples} samples from {n_features} to {n_components} in "
+        f"{time() - t0:0.3f}s"
     )
     if hasattr(rp, "components_"):
         n_bytes = rp.components_.data.nbytes
         n_bytes += rp.components_.indices.nbytes
-        print("Random matrix with size: %0.3fMB" % (n_bytes / 1e6))
+        print(f"Random matrix with size: {n_bytes / 1e6:0.3f} MB")
 
     projected_dists = euclidean_distances(projected_data, squared=True).ravel()[nonzero]
 
@@ -168,7 +168,7 @@
     cb.set_label("Sample pairs counts")
 
     rates = projected_dists / dists
-    print("Mean distances rate: %0.2f (%0.2f)" % (np.mean(rates), np.std(rates)))
+    print(f"Mean distances rate: {np.mean(rates):.2f} ({np.std(rates):.2f})")
 
     plt.figure()
     plt.hist(rates, bins=50, range=(0.0, 2.0), edgecolor="k", density=True)
@@ -186,15 +186,13 @@
 # We can see that for low values of ``n_components`` the distribution is wide
 # with many distorted pairs and a skewed distribution (due to the hard
 # limit of zero ratio on the left as distances are always positives)
-# while for larger values of n_components the distortion is controlled
+# while for larger values of `n_components` the distortion is controlled
 # and the distances are well preserved by the random projection.
-
-
-# %%
+#
 # Remarks
 # =======
 #
-# According to the JL lemma, projecting 500 samples without too much distortion
+# According to the JL lemma, projecting 300 samples without too much distortion
 # will require at least several thousands dimensions, irrespective of the
 # number of features of the original dataset.
 #
@@ -203,5 +201,5 @@
 # for dimensionality reduction in this case.
 #
 # On the twenty newsgroups on the other hand the dimensionality can be
-# decreased from 56436 down to 10000 while reasonably preserving
+# decreased from 56,436 down to 10,000 while reasonably preserving
 # pairwise distances.