WIP: supervised learning, data views

.gitignore

@@ -5,4 +5,5 @@ deps/deps.jl
 
 *.ipynb_checkpoints
 
-Manifest.toml
+/Manifest.toml
+/*.png

NOTICE

@@ -0,0 +1,32 @@
+Contains code modified from umap (https://raw.githubusercontent.com/lmcinnes/umap/)
+which is available under the following License.
+
+BSD 3-Clause License
+
+Copyright (c) 2017, Leland McInnes
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+
+* Neither the name of the copyright holder nor the names of its
+  contributors may be used to endorse or promote products derived from
+  this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

examples/MNIST_demo.jl

@@ -0,0 +1,45 @@
+using UMAP
+using MLDatasets
+using PyCall
+const py_umap = pyimport_conda("umap", "umap-learn")
+
+include("plotting.jl")
+
+# First, let's get a small sample of the MNIST data.
+n_points = 10_000
+X = reshape(MNIST.traintensor(Float64)[:, :, 1:n_points], 28^2, :)
+
+# We'll also get the labels
+y = MNIST.trainlabels()[1:n_points]
+
+# Now we run UMAP super and unsupervised, via PyCall and via UMAP.jl
+
+## try to match py_umap's number of nndescent iterations
+nndescent_kwargs = (max_iters = max(5, round(Int, log2(n_points))), sample_rate = 1)
+unsup = UMAP_(X; n_neighbors=10, min_dist=0.001, n_epochs=200,
+              nndescent_kwargs = nndescent_kwargs)
+py_unsup = py_umap.UMAP(min_dist=0.001, n_epochs=200, n_neighbors=10).fit(permutedims(X))
+
+scene = plot_umap_comparison((permutedims(py_unsup.embedding_), y), (unsup.embedding, y);
+                             titles=("PyUMAP (unsupervised)", "UMAP.jl (unsupervised)"))
+save("MNIST_py_vs_jl_unsupervised.png", scene, px_per_unit=3, resolution=(1440, 810))
+
+
+sup = UMAP_(X, y; n_neighbors=10, min_dist=0.001, n_epochs=200, far_dist=5.0,
+        nndescent_kwargs = nndescent_kwargs)
+py_sup = py_umap.UMAP(min_dist=0.001, n_epochs=200, n_neighbors=10).fit(permutedims(X), y)
+
+scene = plot_umap_comparison((permutedims(py_sup.embedding_), y), (sup.embedding, y);
+                             titles=("PyUMAP (supervised)", "UMAP.jl (supervised)"))
+save("MNIST_py_vs_jl_supervised.png", scene, px_per_unit=3, resolution=(1440, 810))
+
+
+scene = plot_umap_comparison((unsup.embedding, y), (sup.embedding, y);
+                             titles=("UMAP.jl (unsupervised)", "UMAP.jl (supervised)"))
+save("MNIST_jl_unsup_vs_supervised.png", scene, px_per_unit=3, resolution=(1440, 810))
+
+
+scene = plot_umap_comparison((permutedims(py_unsup.embedding_), y),
+                             (permutedims(py_sup.embedding_), y);
+                             titles=("PyUMAP (unsupervised)", "PyUMAP (supervised)"))
+save("MNIST_py_unsup_vs_supervised.png", scene, px_per_unit=3, resolution=(1440, 810))

examples/MNIST_no_python.jl

@@ -0,0 +1,50 @@
+using UMAP
+using MLDatasets
+using Distances
+
+include("plotting.jl")
+
+# First, let's get a small sample of the MNIST data.
+n_points = 1_000
+X = reshape(MNIST.traintensor(Float64)[:, :, 1:n_points], 28^2, :)
+
+# We'll also get the labels
+y = MNIST.trainlabels()[1:n_points]
+
+# Now we run UMAP super and unsupervised, via PyCall and via UMAP.jl
+
+## try to match py_umap's number of nndescent iterations
+nndescent_kwargs = (max_iters = max(5, round(Int, log2(n_points))), sample_rate = 1)
+n_neighbors = 20
+n_neighbors_meta = 200
+
+unsup = Embedding(DataView(X; metric=Euclidean(), n_neighbors, nndescent_kwargs);  min_dist=0.001, n_epochs=200)
+
+sup = Embedding(DataView(X; metric=Euclidean(), n_neighbors, nndescent_kwargs), DataView(y;metric=Categorical(far_dist=5.0), n_neighbors=n_neighbors_meta, nndescent_kwargs);  min_dist=0.001, n_epochs=200)
+
+# Bug in NearestNeighbors or Distances?
+# NN calls `result_type(metric, data[1], data[2])` instead of `result_type(metric, typeof(data[1]), typeof(data[2])`.
+# But this works for arrays...
+Distances.result_type(M, ::Int, ::Int) = Distances.result_type(M, Int, Int)
+#
+let
+    scene = plot_umap_comparison((unsup.embedding, y), (sup.embedding, y);
+        titles=("UMAP.jl (unsupervised)", "UMAP.jl (supervised)"), markersize=5px)
+    save("MNIST_jl_unsup_vs_supervised.png", scene, px_per_unit=3, resolution=(1440, 810))
+end
+
+
+
+for mix_weight in (0.001, 0.25, 0.5, 0.75, 0.999), n_neighbors_meta in (20, 200)
+    sup_cts = Embedding(DataView(X; metric=Euclidean(), n_neighbors, nndescent_kwargs), DataView(y;metric=Euclidean(), n_neighbors=n_neighbors_meta, nndescent_kwargs);  min_dist=0.001, n_epochs=200, mix_weight)
+
+    scene = plot_umap_comparison((sup.embedding, y), (sup_cts.embedding, y);
+        titles=("UMAP.jl (supervised; categorical)", "UMAP.jl (supervised; continuous) $(n_neighbors_meta) neighbors), mix_weight=$(mix_weight)"), markersize=5px)
+    save("MNIST_jl_vs_jl_supervised_cat_vs_cts_max_weight_$(mix_weight)_neighbors_$(n_neighbors_meta).png", scene, px_per_unit=3, resolution=(1440, 810))
+
+
+    scene = plot_umap_comparison((unsup.embedding, y), (sup_cts.embedding, y);
+    titles=("UMAP.jl (unsupervised)", "UMAP.jl (supervised; continuous) $(n_neighbors_meta) neighbors), mix_weight=$(mix_weight)"), markersize=5px)
+save("MNIST_jl_vs_jl_unsupervised_vs_cts_max_weight_$(mix_weight)_neighbors_$(n_neighbors_meta).png", scene, px_per_unit=3, resolution=(1440, 810))
+
+end