.. DO NOT EDIT. .. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY. .. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE: .. "auto_examples/basics/demo_ne_methods_affinity_matcher.py" .. LINE NUMBERS ARE GIVEN BELOW. .. only:: html .. note:: :class: sphx-glr-download-link-note :ref:`Go to the end ` to download the full example code. .. rst-class:: sphx-glr-example-title .. _sphx_glr_auto_examples_basics_demo_ne_methods_affinity_matcher.py: Neighbor Embedding on genomics & equivalent affinity matcher formulation ========================================================================= We illustrate the basic usage of TorchDR with different neighbor embedding methods on the SNARE-seq gene expression dataset with given cell type labels. .. GENERATED FROM PYTHON SOURCE LINES 9-17 .. code-block:: Python # Author: Titouan Vayer # Hugues Van Assel # # License: BSD 3-Clause License import urllib.request .. GENERATED FROM PYTHON SOURCE LINES 18-30 .. code-block:: Python import matplotlib.pyplot as plt import numpy as np from torchdr import ( SNE, TSNE, UMAP, AffinityMatcher, EntropicAffinity, NormalizedGaussianAffinity, ) .. GENERATED FROM PYTHON SOURCE LINES 31-33 Load the SNARE-seq dataset (gene expression) with cell type labels ------------------------------------------------------------------- .. GENERATED FROM PYTHON SOURCE LINES 33-50 .. code-block:: Python def load_numpy_from_url(url, delimiter="\t"): response = urllib.request.urlopen(url) data = response.read().decode("utf-8") data = data.split("\n") data = [row.split(delimiter) for row in data if row] numpy_array = np.array(data, dtype=float) return numpy_array url_x = "https://rsinghlab.github.io/SCOT/data/snare_rna.txt" X = load_numpy_from_url(url_x) url_y = "https://rsinghlab.github.io/SCOT/data/SNAREseq_types.txt" Y = load_numpy_from_url(url_y) .. GENERATED FROM PYTHON SOURCE LINES 51-53 Run neighbor embedding methods ------------------------------- .. GENERATED FROM PYTHON SOURCE LINES 53-77 .. code-block:: Python params = { "optimizer": "Adam", "optimizer_kwargs": None, "max_iter": 100, "lr": 1e0, } sne = SNE(early_exaggeration_coeff=1, **params) umap = UMAP(early_exaggeration_coeff=1, **params) tsne = TSNE(early_exaggeration_coeff=1, **params) all_methods = { "TSNE": tsne, "SNE": sne, "UMAP": umap, } for method_name, method in all_methods.items(): print(f"--- Computing {method_name} ---") method.fit(X) .. rst-class:: sphx-glr-script-out .. code-block:: none --- Computing TSNE --- --- Computing SNE --- --- Computing UMAP --- .. GENERATED FROM PYTHON SOURCE LINES 78-80 Plot the different embeddings ----------------------------- .. GENERATED FROM PYTHON SOURCE LINES 80-92 .. code-block:: Python fig = plt.figure(figsize=(12, 4)) for i, (method_name, method) in enumerate(all_methods.items()): ax = fig.add_subplot(1, 3, i + 1) emb = method.embedding_.detach().numpy() # get the embedding ax.scatter(emb[:, 0], emb[:, 1], c=Y, s=10) ax.set_title(f"{method_name}", fontsize=24) ax.set_xticks([]) ax.set_yticks([]) plt.tight_layout() .. image-sg:: /auto_examples/basics/images/sphx_glr_demo_ne_methods_affinity_matcher_001.png :alt: TSNE, SNE, UMAP :srcset: /auto_examples/basics/images/sphx_glr_demo_ne_methods_affinity_matcher_001.png :class: sphx-glr-single-img .. GENERATED FROM PYTHON SOURCE LINES 93-104 Using AffinityMatcher ----------------------------- We can reproduce the same embeddings using the class :class:`torchdr.AffinityMatcher`. The latter takes as input two affinities and minimize a certain matching loss between them. To reproduce the SNE algorithm we can match, via the cross entropy loss, a :class:`torchdr.EntropicAffinity` with a :class:`torchdr.NormalizedGaussianAffinity`. .. GENERATED FROM PYTHON SOURCE LINES 104-127 .. code-block:: Python sne_affinity_matcher = AffinityMatcher( n_components=2, # SNE matches an EntropicAffinity affinity_in=EntropicAffinity(sparsity=False), # with a Gaussian kernel normalized by row affinity_out=NormalizedGaussianAffinity(normalization_dim=1), loss_fn="cross_entropy_loss", # and the cross_entropy loss **params, ) sne_affinity_matcher.fit(X) fig = plt.figure(figsize=(8, 4)) two_sne_dict = {"SNE": sne, "SNE (with affinity matcher)": sne_affinity_matcher} for i, (method_name, method) in enumerate(two_sne_dict.items()): ax = fig.add_subplot(1, 2, i + 1) emb = method.embedding_.detach().numpy() # get the embedding ax.scatter(emb[:, 0], emb[:, 1], c=Y, s=10) ax.set_title(f"{method_name}", fontsize=15) ax.set_xticks([]) ax.set_yticks([]) plt.tight_layout() .. image-sg:: /auto_examples/basics/images/sphx_glr_demo_ne_methods_affinity_matcher_002.png :alt: SNE, SNE (with affinity matcher) :srcset: /auto_examples/basics/images/sphx_glr_demo_ne_methods_affinity_matcher_002.png :class: sphx-glr-single-img .. GENERATED FROM PYTHON SOURCE LINES 128-138 On the efficiency of using the torchdr API rather than AffinityMatcher directly ------------------------------------------------------------------------------- .. note:: Calling :class:`torchdr.SNE` enables to leverage sparsity and therefore significantly reduces the computational cost of the algorithm compared to using :class:`torchdr.AffinityMatcher` with the corresponding affinities. In TorchDR, it is therefore recommended to use the specific class associated with the desired algorithm when available. .. rst-class:: sphx-glr-timing **Total running time of the script:** (1 minutes 3.481 seconds) .. _sphx_glr_download_auto_examples_basics_demo_ne_methods_affinity_matcher.py: .. only:: html .. container:: sphx-glr-footer sphx-glr-footer-example .. container:: sphx-glr-download sphx-glr-download-jupyter :download:`Download Jupyter notebook: demo_ne_methods_affinity_matcher.ipynb ` .. container:: sphx-glr-download sphx-glr-download-python :download:`Download Python source code: demo_ne_methods_affinity_matcher.py ` .. container:: sphx-glr-download sphx-glr-download-zip :download:`Download zipped: demo_ne_methods_affinity_matcher.zip ` .. only:: html .. rst-class:: sphx-glr-signature `Gallery generated by Sphinx-Gallery `_