<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta name="generator" content="Docutils 0.18.1: http://docutils.sourceforge.net/" />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Multiple-Instance Learning (MIL) — slideflow 3.0.0 documentation</title>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="../_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="index" title="Index" href="../genindex/" />
<link rel="search" title="Search" href="../search/" />
<link rel="next" title="Self-Supervised Learning (SSL)" href="../ssl/" />
<link rel="prev" title="Generating Features" href="../features/" />
<script src="../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
<script defer data-domain="slideflow.dev" src="https://plausible.io/js/script.js"></script>
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://slideflow.dev" aria-label="Slideflow"></a>
<div class="main-menu">
<ul>
<li class="active">
<a href="https://slideflow.dev">Docs</a>
</li>
<li>
<a href="https://slideflow.dev/tutorial1/">Tutorials</a>
</li>
<li>
<a href="https://github.com/slideflow/slideflow">GitHub</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
3.0
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search/" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption" role="heading"><span class="caption-text">Introduction</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="../installation/">Installation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../overview/">Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../quickstart/">Quickstart</a></li>
<li class="toctree-l1"><a class="reference internal" href="../project_setup/">Setting up a Project</a></li>
<li class="toctree-l1"><a class="reference internal" href="../datasets_and_val/">Datasets</a></li>
<li class="toctree-l1"><a class="reference internal" href="../slide_processing/">Slide Processing</a></li>
<li class="toctree-l1"><a class="reference internal" href="../training/">Training</a></li>
<li class="toctree-l1"><a class="reference internal" href="../evaluation/">Evaluation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../posthoc/">Layer Activations</a></li>
<li class="toctree-l1"><a class="reference internal" href="../uq/">Uncertainty Quantification</a></li>
<li class="toctree-l1"><a class="reference internal" href="../features/">Generating Features</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Multiple-Instance Learning (MIL)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../ssl/">Self-Supervised Learning (SSL)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../stylegan/">Generative Networks (GANs)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../saliency/">Saliency Maps</a></li>
<li class="toctree-l1"><a class="reference internal" href="../segmentation/">Tissue Segmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../cellseg/">Cell Segmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../custom_loops/">Custom Training Loops</a></li>
<li class="toctree-l1"><a class="reference internal" href="../studio/">Slideflow Studio: Live Visualization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../troubleshooting/">Troubleshooting</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../tfrecords/">TFRecords: Reading and Writing</a></li>
<li class="toctree-l1"><a class="reference internal" href="../dataloaders/">Dataloaders: Sampling and Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../custom_extractors/">Custom Feature Extractors</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tile_labels/">Strong Supervision with Tile Labels</a></li>
<li class="toctree-l1"><a class="reference internal" href="../plugins/">Creating a Slideflow Plugin</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../slideflow/">slideflow</a></li>
<li class="toctree-l1"><a class="reference internal" href="../project/">slideflow.Project</a></li>
<li class="toctree-l1"><a class="reference internal" href="../dataset/">slideflow.Dataset</a></li>
<li class="toctree-l1"><a class="reference internal" href="../dataset_features/">slideflow.DatasetFeatures</a></li>
<li class="toctree-l1"><a class="reference internal" href="../heatmap/">slideflow.Heatmap</a></li>
<li class="toctree-l1"><a class="reference internal" href="../model_params/">slideflow.ModelParams</a></li>
<li class="toctree-l1"><a class="reference internal" href="../mosaic/">slideflow.Mosaic</a></li>
<li class="toctree-l1"><a class="reference internal" href="../slidemap/">slideflow.SlideMap</a></li>
<li class="toctree-l1"><a class="reference internal" href="../biscuit/">slideflow.biscuit</a></li>
<li class="toctree-l1"><a class="reference internal" href="../slideflow_cellseg/">slideflow.cellseg</a></li>
<li class="toctree-l1"><a class="reference internal" href="../io/">slideflow.io</a></li>
<li class="toctree-l1"><a class="reference internal" href="../io_tensorflow/">slideflow.io.tensorflow</a></li>
<li class="toctree-l1"><a class="reference internal" href="../io_torch/">slideflow.io.torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../gan/">slideflow.gan</a></li>
<li class="toctree-l1"><a class="reference internal" href="../grad/">slideflow.grad</a></li>
<li class="toctree-l1"><a class="reference internal" href="../mil_module/">slideflow.mil</a></li>
<li class="toctree-l1"><a class="reference internal" href="../model/">slideflow.model</a></li>
<li class="toctree-l1"><a class="reference internal" href="../model_tensorflow/">slideflow.model.tensorflow</a></li>
<li class="toctree-l1"><a class="reference internal" href="../model_torch/">slideflow.model.torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../norm/">slideflow.norm</a></li>
<li class="toctree-l1"><a class="reference internal" href="../simclr/">slideflow.simclr</a></li>
<li class="toctree-l1"><a class="reference internal" href="../slide/">slideflow.slide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../slide_qc/">slideflow.slide.qc</a></li>
<li class="toctree-l1"><a class="reference internal" href="../stats/">slideflow.stats</a></li>
<li class="toctree-l1"><a class="reference internal" href="../util/">slideflow.util</a></li>
<li class="toctree-l1"><a class="reference internal" href="../studio_module/">slideflow.studio</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Tutorials</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../tutorial1/">Tutorial 1: Model training (simple)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial2/">Tutorial 2: Model training (advanced)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial3/">Tutorial 3: Using a custom architecture</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial4/">Tutorial 4: Model evaluation & heatmaps</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial5/">Tutorial 5: Creating a mosaic map</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial6/">Tutorial 6: Custom slide filtering</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial7/">Tutorial 7: Training with custom augmentations</a></li>
<li class="toctree-l1"><a class="reference internal" href="../tutorial8/">Tutorial 8: Multiple-Instance Learning</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="../">
Docs
</a> >
</li>
<li>Multiple-Instance Learning (MIL)</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/mil.rst.txt" rel="nofollow"><img src="../_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<section id="multiple-instance-learning-mil">
<span id="mil"></span><h1>Multiple-Instance Learning (MIL)<a class="headerlink" href="#multiple-instance-learning-mil" title="Permalink to this heading">¶</a></h1>
<p>In addition to standard tile-based neural networks, Slideflow also supports training multiple-instance learning (MIL) models. Several architectures are available, including <a class="reference external" href="https://github.com/AMLab-Amsterdam/AttentionDeepMIL">attention-based MIL</a> (<code class="docutils literal notranslate"><span class="pre">"Attention_MIL"</span></code>), <a class="reference external" href="https://github.com/mahmoodlab/CLAM">CLAM</a> (<code class="docutils literal notranslate"><span class="pre">"CLAM_SB",</span></code> <code class="docutils literal notranslate"><span class="pre">"CLAM_MB"</span></code>, <code class="docutils literal notranslate"><span class="pre">"MIL_fc"</span></code>, <code class="docutils literal notranslate"><span class="pre">"MIL_fc_mc"</span></code>), <a class="reference external" href="https://github.com/szc19990412/TransMIL">TransMIL</a> (<code class="docutils literal notranslate"><span class="pre">"TransMIL"</span></code>), and <a class="reference external" href="https://github.com/peng-lab/HistoBistro">HistoBistro Transformer</a> (<code class="docutils literal notranslate"><span class="pre">"bistro.transformer"</span></code>). Custom architectures can also be trained. MIL training requires PyTorch.</p>
<p>Skip to <a class="reference internal" href="../tutorial8/#tutorial8"><span class="std std-ref">Tutorial 8: Multiple-Instance Learning</span></a> for a complete example of MIL training.</p>
<p>See <a class="reference internal" href="../mil_module/#mil-api"><span class="std std-ref">slideflow.mil</span></a> for more information on the MIL API.</p>
<section id="generating-features">
<h2>Generating Features<a class="headerlink" href="#generating-features" title="Permalink to this heading">¶</a></h2>
<p>The first step in MIL model development is generating features from image tiles, as discussed in the <a class="reference internal" href="../features/#features"><span class="std std-ref">Generating Features</span></a> section. Features from whole-slide images are exported as “bags” of features, where each bag contains a set of features from a single slide. Each bag is a PyTorch tensor saved in <code class="docutils literal notranslate"><span class="pre">*.pt</span></code> format. Bags are saved in a directory, and the directory path is passed to the MIL model during training and evaluation.</p>
</section>
<section id="training">
<h2>Training<a class="headerlink" href="#training" title="Permalink to this heading">¶</a></h2>
<section id="model-configuration">
<h3>Model Configuration<a class="headerlink" href="#model-configuration" title="Permalink to this heading">¶</a></h3>
<p>To train an MIL model using exported features, first prepare an MIL configuration using <a class="reference internal" href="../mil_module/#slideflow.mil.mil_config" title="slideflow.mil.mil_config"><code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.mil.mil_config()</span></code></a>.</p>
<p>The first argument to this function is the model architecture (which can be a name or a custom <code class="docutils literal notranslate"><span class="pre">torch.nn.Module</span></code> model), and the remaining arguments are used to configure the training process, such as learning rate and number of epochs. Training is executed using <a class="reference external" href="https://docs.fast.ai/">FastAI</a> with <a class="reference external" href="https://arxiv.org/pdf/1803.09820.pdf%E5%92%8CSylvain">1cycle learning rate scheduling</a>.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">slideflow</span> <span class="k">as</span> <span class="nn">sf</span>
<span class="kn">from</span> <span class="nn">slideflow.mil</span> <span class="kn">import</span> <span class="n">mil_config</span>
<span class="n">config</span> <span class="o">=</span> <span class="n">mil_config</span><span class="p">(</span><span class="s1">'attention_mil'</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">)</span>
</pre></div>
</div>
<p>Available models out-of-the-box include <a class="reference external" href="https://github.com/AMLab-Amsterdam/AttentionDeepMIL">attention-based MIL</a> (<code class="docutils literal notranslate"><span class="pre">"Attention_MIL"</span></code>), <a class="reference external" href="https://github.com/szc19990412/TransMIL">transformer MIL</a> (<code class="docutils literal notranslate"><span class="pre">"TransMIL"</span></code>), and <a class="reference external" href="https://github.com/peng-lab/HistoBistro">HistoBistro Transformer</a> (<code class="docutils literal notranslate"><span class="pre">"bistro.transformer"</span></code>). <a class="reference external" href="https://github.com/mahmoodlab/CLAM">CLAM</a> (<code class="docutils literal notranslate"><span class="pre">"CLAM_SB",</span></code> <code class="docutils literal notranslate"><span class="pre">"CLAM_MB"</span></code>, <code class="docutils literal notranslate"><span class="pre">"MIL_fc"</span></code>, <code class="docutils literal notranslate"><span class="pre">"MIL_fc_mc"</span></code>) models are available through <code class="docutils literal notranslate"><span class="pre">slideflow-gpl</span></code>:</p>
<div class="highlight-bash notranslate"><div class="highlight"><pre><span></span>pip<span class="w"> </span>install<span class="w"> </span>slideflow-gpl
</pre></div>
</div>
<p>Custom MIL models can also be trained with this API, as discussed <a class="reference external" href="#custom_mil">below</a>.</p>
</section>
<section id="classification-regression">
<h3>Classification & Regression<a class="headerlink" href="#classification-regression" title="Permalink to this heading">¶</a></h3>
<p>MIL models can be trained for both classification and regression tasks. The type of outcome is determined through the loss function, which defaults to <code class="docutils literal notranslate"><span class="pre">"cross_entropy"</span></code>. To train a model for regression, set the loss function to one of the following regression losses, and ensure that your outcome labels are continuous. You can also train to multiple outcomes by passing a list of outcome names.</p>
<ul class="simple">
<li><p><strong>“mse”</strong> (<code class="docutils literal notranslate"><span class="pre">nn.CrossEntropyLoss</span></code>): Mean squared error.</p></li>
<li><p><strong>“mae”</strong> (<code class="docutils literal notranslate"><span class="pre">nn.L1Loss</span></code>): Mean absolute error.</p></li>
<li><p><strong>“huber”</strong> (<code class="docutils literal notranslate"><span class="pre">nn.SmoothL1Loss</span></code>): Huber loss.</p></li>
</ul>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Prepare a regression-compatible MIL configuration</span>
<span class="n">config</span> <span class="o">=</span> <span class="n">mil_config</span><span class="p">(</span><span class="s1">'attention_mil'</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">,</span> <span class="n">loss</span><span class="o">=</span><span class="s1">'mse'</span><span class="p">)</span>
<span class="c1"># Train the model</span>
<span class="n">project</span><span class="o">.</span><span class="n">train_mil</span><span class="p">(</span>
<span class="n">config</span><span class="o">=</span><span class="n">config</span><span class="p">,</span>
<span class="o">...</span><span class="p">,</span>
<span class="n">outcomes</span><span class="o">=</span><span class="p">[</span><span class="s1">'age'</span><span class="p">,</span> <span class="s1">'grade'</span><span class="p">]</span>
<span class="p">)</span>
</pre></div>
</div>
</section>
<section id="training-an-mil-model">
<h3>Training an MIL Model<a class="headerlink" href="#training-an-mil-model" title="Permalink to this heading">¶</a></h3>
<p>Next, prepare a <a class="reference internal" href="../datasets_and_val/#datasets-and-validation"><span class="std std-ref">training and validation dataset</span></a> and use <a class="reference internal" href="../project/#slideflow.Project.train_mil" title="slideflow.Project.train_mil"><code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.Project.train_mil()</span></code></a> to start training. For example, to train a model using three-fold cross-validation to the outcome “HPV_status”:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="o">...</span>
<span class="c1"># Prepare a project and dataset</span>
<span class="n">P</span> <span class="o">=</span> <span class="n">sf</span><span class="o">.</span><span class="n">Project</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
<span class="n">full_dataset</span> <span class="o">=</span> <span class="n">dataset</span> <span class="o">=</span> <span class="n">P</span><span class="o">.</span><span class="n">dataset</span><span class="p">(</span><span class="n">tile_px</span><span class="o">=</span><span class="mi">299</span><span class="p">,</span> <span class="n">tile_um</span><span class="o">=</span><span class="mi">302</span><span class="p">)</span>
<span class="c1"># Split the dataset using three-fold, site-preserved cross-validation</span>
<span class="n">splits</span> <span class="o">=</span> <span class="n">full_dataset</span><span class="o">.</span><span class="n">kfold_split</span><span class="p">(</span>
<span class="n">k</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span>
<span class="n">labels</span><span class="o">=</span><span class="s1">'HPV_status'</span><span class="p">,</span>
<span class="n">preserved_site</span><span class="o">=</span><span class="kc">True</span>
<span class="p">)</span>
<span class="c1"># Train on each cross-fold</span>
<span class="k">for</span> <span class="n">train</span><span class="p">,</span> <span class="n">val</span> <span class="ow">in</span> <span class="n">splits</span><span class="p">:</span>
<span class="n">P</span><span class="o">.</span><span class="n">train_mil</span><span class="p">(</span>
<span class="n">config</span><span class="o">=</span><span class="n">config</span><span class="p">,</span>
<span class="n">outcomes</span><span class="o">=</span><span class="s1">'HPV_status'</span><span class="p">,</span>
<span class="n">train_dataset</span><span class="o">=</span><span class="n">train</span><span class="p">,</span>
<span class="n">val_dataset</span><span class="o">=</span><span class="n">val</span><span class="p">,</span>
<span class="n">bags</span><span class="o">=</span><span class="s1">'/path/to/bag_directory'</span>
<span class="p">)</span>
</pre></div>
</div>
<p>Model training statistics, including validation performance (AUROC, AP) and predictions on the validation dataset, will be saved in an <code class="docutils literal notranslate"><span class="pre">mil</span></code> subfolder within the main project directory.</p>
<p>If you are training an attention-based MIL model (<code class="docutils literal notranslate"><span class="pre">attention_mil</span></code>, <code class="docutils literal notranslate"><span class="pre">clam_sb</span></code>, <code class="docutils literal notranslate"><span class="pre">clam_mb</span></code>), heatmaps of attention can be generated for each slide in the validation dataset by using the argument <code class="docutils literal notranslate"><span class="pre">attention_heatmaps=True</span></code>. You can customize these heatmaps with <code class="docutils literal notranslate"><span class="pre">interpolation</span></code> and <code class="docutils literal notranslate"><span class="pre">cmap</span></code> arguments to control the heatmap interpolation and colormap, respectively.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Generate attention heatmaps,</span>
<span class="c1"># using the 'magma' colormap and no interpolation.</span>
<span class="n">P</span><span class="o">.</span><span class="n">train_mil</span><span class="p">(</span>
<span class="n">attention_heatmaps</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">cmap</span><span class="o">=</span><span class="s1">'magma'</span><span class="p">,</span>
<span class="n">interpolation</span><span class="o">=</span><span class="kc">None</span>
<span class="p">)</span>
</pre></div>
</div>
<p>Hyperparameters, model configuration, and feature extractor information is logged to <code class="docutils literal notranslate"><span class="pre">mil_params.json</span></code> in the model directory. This file also contains information about the input and output shapes of the MIL network and outcome labels. An example file is shown below.</p>
<div class="highlight-json notranslate"><div class="highlight"><pre><span></span><span class="p">{</span>
<span class="w"> </span><span class="nt">"trainer"</span><span class="p">:</span><span class="w"> </span><span class="s2">"fastai"</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"params"</span><span class="p">:</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="p">},</span>
<span class="w"> </span><span class="nt">"outcomes"</span><span class="p">:</span><span class="w"> </span><span class="s2">"histology"</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"outcome_labels"</span><span class="p">:</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="nt">"0"</span><span class="p">:</span><span class="w"> </span><span class="s2">"Adenocarcinoma"</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"1"</span><span class="p">:</span><span class="w"> </span><span class="s2">"Squamous"</span>
<span class="w"> </span><span class="p">},</span>
<span class="w"> </span><span class="nt">"bags"</span><span class="p">:</span><span class="w"> </span><span class="s2">"/mnt/data/projects/example_project/bags/simclr-263510/"</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"input_shape"</span><span class="p">:</span><span class="w"> </span><span class="mi">1024</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"output_shape"</span><span class="p">:</span><span class="w"> </span><span class="mi">2</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"bags_encoder"</span><span class="p">:</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="nt">"extractor"</span><span class="p">:</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="nt">"class"</span><span class="p">:</span><span class="w"> </span><span class="s2">"slideflow.model.extractors.simclr.SimCLR_Features"</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"kwargs"</span><span class="p">:</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="nt">"center_crop"</span><span class="p">:</span><span class="w"> </span><span class="kc">false</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"ckpt"</span><span class="p">:</span><span class="w"> </span><span class="s2">"/mnt/data/projects/example_project/simclr/00001-EXAMPLE/ckpt-263510.ckpt"</span>
<span class="w"> </span><span class="p">}</span>
<span class="w"> </span><span class="p">},</span>
<span class="w"> </span><span class="nt">"normalizer"</span><span class="p">:</span><span class="w"> </span><span class="kc">null</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"num_features"</span><span class="p">:</span><span class="w"> </span><span class="mi">1024</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"tile_px"</span><span class="p">:</span><span class="w"> </span><span class="mi">299</span><span class="p">,</span>
<span class="w"> </span><span class="nt">"tile_um"</span><span class="p">:</span><span class="w"> </span><span class="mi">302</span>
<span class="w"> </span><span class="p">}</span>
<span class="p">}</span>
</pre></div>
</div>
</section>
<section id="multi-magnification-mil">
<span id="multimag"></span><h3>Multi-Magnification MIL<a class="headerlink" href="#multi-magnification-mil" title="Permalink to this heading">¶</a></h3>
<p>Slideflow 2.2 introduced a multi-magnification, multi-modal MIL model, <code class="docutils literal notranslate"><span class="pre">MultiModal_Attention_MIL</span></code> (<code class="docutils literal notranslate"><span class="pre">"mm_attention_mil"</span></code>). This late-fusion multimodal model is based on standard attention-based MIL, but accepts multiple input modalities (e.g., multiple magnifications) simultaneously. Each input modality is processed by a separate encoder network and a separate attention module. The attention-weighted features from each modality are then concatenated and passed to a fully-connected layer.</p>
<p>Multimodal models are trained using the same API as standard MIL models. Modalities are specified using the <code class="docutils literal notranslate"><span class="pre">bags</span></code> argument to <a class="reference internal" href="../project/#slideflow.Project.train_mil" title="slideflow.Project.train_mil"><code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.Project.train_mil()</span></code></a>, where the number of modes is determined by the number of bag directories provided. Within each bag directory, bags should be generated using the same feature extractor and at the same magnification, but feature extractors and magnifications can vary between bag directories.</p>
<p>For example, to train a multimodal model using two magnifications, you would pass two bag paths to the model. In this case, the <code class="docutils literal notranslate"><span class="pre">/path/to/bags_10x</span></code> directory contains bags generated from a 10x feature extractor, and the <code class="docutils literal notranslate"><span class="pre">/path/to/bags_40x</span></code> directory contains bags generated from a 40x feature extractor.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="c1"># Configure a multimodal MIL model.</span>
<span class="n">config</span> <span class="o">=</span> <span class="n">mil_config</span><span class="p">(</span><span class="s1">'mm_attention_mil'</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-4</span><span class="p">)</span>
<span class="c1"># Set the bags paths for each modality.</span>
<span class="n">bags_10x</span> <span class="o">=</span> <span class="s1">'/path/to/bags_10x'</span>
<span class="n">bags_40x</span> <span class="o">=</span> <span class="s1">'/path/to/bags_40x'</span>
<span class="n">P</span><span class="o">.</span><span class="n">train_mil</span><span class="p">(</span>
<span class="n">config</span><span class="o">=</span><span class="n">config</span><span class="p">,</span>
<span class="n">outcomes</span><span class="o">=</span><span class="s1">'HPV_status'</span><span class="p">,</span>
<span class="n">train_dataset</span><span class="o">=</span><span class="n">train</span><span class="p">,</span>
<span class="n">val_dataset</span><span class="o">=</span><span class="n">val</span><span class="p">,</span>
<span class="n">bags</span><span class="o">=</span><span class="p">[</span><span class="n">bags_10x</span><span class="p">,</span> <span class="n">bags_40x</span><span class="p">]</span>
<span class="p">)</span>
</pre></div>
</div>
<p>You can use any number of modalities, and the feature extractors for each modality can be different. For example, you could train a multimodal model using features from a custom SimCLR model at 5x and features from a pretrained CTransPath model at 20x.</p>
<p>The feature extractors used for each modality, as specified in the <code class="docutils literal notranslate"><span class="pre">bags_config.json</span></code> files in the bag directories, will be logged in the final <code class="docutils literal notranslate"><span class="pre">mil_params.json</span></code> file. Multimodal MIL models can be interactively viewed in <a class="reference internal" href="../studio/#studio"><span class="std std-ref">Slideflow Studio</span></a>, allowing you to visualize the attention weights for each modality separately.</p>
</section>
<section id="custom-architectures">
<span id="custom-mil"></span><h3>Custom Architectures<a class="headerlink" href="#custom-architectures" title="Permalink to this heading">¶</a></h3>
<p>Training custom MIL models is straightforward with Slideflow, particularly if your model can adhere to a few simple guidelines:</p>
<ul class="simple">
<li><p>Initialized with <code class="docutils literal notranslate"><span class="pre">(num_feats,</span> <span class="pre">num_outputs)</span></code> (e.g., <code class="docutils literal notranslate"><span class="pre">Attention_MIL(768,</span> <span class="pre">2)</span></code>)</p></li>
<li><p>Input is feature bags with shape <code class="docutils literal notranslate"><span class="pre">(batch,</span> <span class="pre">num_tiles,</span> <span class="pre">num_feats)</span></code>. If the model needs a “lens” input, then the model attribute <code class="docutils literal notranslate"><span class="pre">use_lens</span></code> should be True.</p></li>
<li><p>Has a <code class="docutils literal notranslate"><span class="pre">relocate()</span></code> function that moves the model to detected device/GPU</p></li>
<li><dl class="simple">
<dt>Ability to get attention through one of two methods:</dt><dd><ul>
<li><p><code class="docutils literal notranslate"><span class="pre">forward()</span></code> function includes an optional <code class="docutils literal notranslate"><span class="pre">return_attention</span></code> argument, which if True returns attention scores after model output</p></li>
<li><p>Has a <code class="docutils literal notranslate"><span class="pre">calculate_attention()</span></code> function that returns attention scores</p></li>
</ul>
</dd>
</dl>
</li>
</ul>
<p>If the above applies to your model, you can train it simply by passing it as the first argument to <a class="reference internal" href="../mil_module/#slideflow.mil.mil_config" title="slideflow.mil.mil_config"><code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.mil.mil_config()</span></code></a>.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">slideflow</span> <span class="k">as</span> <span class="nn">sf</span>
<span class="kn">from</span> <span class="nn">slideflow.mil</span> <span class="kn">import</span> <span class="n">mil_config</span>
<span class="kn">from</span> <span class="nn">my_module</span> <span class="kn">import</span> <span class="n">CustomMIL</span>
<span class="n">config</span> <span class="o">=</span> <span class="n">mil_config</span><span class="p">(</span><span class="n">CustomMIL</span><span class="p">,</span> <span class="n">lr</span><span class="o">=</span><span class="mf">1e-3</span><span class="p">)</span>
</pre></div>
</div>
<p>For larger projects, or if you are designing a plugin/extension for Slideflow, custom models can be registered to facilitate easy creation. If your model adheres to the above guidelines, you can register it for use with the following:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">slideflow.mil</span> <span class="kn">import</span> <span class="n">register_model</span>
<span class="nd">@register_model</span>
<span class="k">def</span> <span class="nf">my_model</span><span class="p">():</span>
<span class="k">return</span> <span class="n">MyModelClass</span>
</pre></div>
</div>
<p>You can then use your model when creating an MIL configuration:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">config</span> <span class="o">=</span> <span class="n">sf</span><span class="o">.</span><span class="n">mil</span><span class="o">.</span><span class="n">mil_config</span><span class="p">(</span><span class="s1">'my_model'</span><span class="p">,</span> <span class="o">...</span><span class="p">)</span>
</pre></div>
</div>
<p>If the above guidelines do <em>not</em> apply to your model, or if you want to customize model logic or functionality, you can supply a custom MIL configuration class that will supervise model building and dataset preparation. Your custom configuration class should inherit <code class="docutils literal notranslate"><span class="pre">slideflow.mil.MILModelConfig</span></code>, and methods in this class can be overloaded to provide additional functionality. For example, to create an MIL configuration that uses a custom loss and custom metrics:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">slideflow.mil</span> <span class="kn">import</span> <span class="n">MILModelConfig</span>
<span class="k">class</span> <span class="nc">MyModelConfig</span><span class="p">(</span><span class="n">MILModelConfig</span><span class="p">):</span>
<span class="nd">@property</span>
<span class="k">def</span> <span class="nf">loss_fn</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="n">my_custom_loss</span>
<span class="k">def</span> <span class="nf">get_metrics</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="p">[</span><span class="n">my_metric1</span><span class="p">,</span> <span class="n">my_metric2</span><span class="p">]</span>
</pre></div>
</div>
<p>When registering your model, you should specify that it should use your custom configuration:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="nd">@register_model</span><span class="p">(</span><span class="n">config</span><span class="o">=</span><span class="n">MyModelConfig</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">my_model</span><span class="p">():</span>
<span class="k">return</span> <span class="n">MyModelClass</span>
</pre></div>
</div>
<p>For an example of how to utilize model registration and configuration customization, see our <a class="reference external" href="https://github.com/slideflow/slideflow-gpl/blob/main/slideflow_gpl/clam/config.py">CLAM implementation</a> available through <code class="docutils literal notranslate"><span class="pre">slideflow-gpl</span></code>.</p>
</section>
</section>
<section id="evaluation">
<h2>Evaluation<a class="headerlink" href="#evaluation" title="Permalink to this heading">¶</a></h2>
<p>To evaluate a saved MIL model on an external dataset, first extract features from a dataset, then use <a class="reference internal" href="../project/#slideflow.Project.evaluate_mil" title="slideflow.Project.evaluate_mil"><code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.Project.evaluate_mil()</span></code></a>, which displays evaluation metrics and returns predictions as a DataFrame.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">slideflow</span> <span class="k">as</span> <span class="nn">sf</span>
<span class="c1"># Prepare a project and dataset</span>
<span class="n">P</span> <span class="o">=</span> <span class="n">sf</span><span class="o">.</span><span class="n">Project</span><span class="p">(</span><span class="o">...</span><span class="p">)</span>
<span class="n">dataset</span> <span class="o">=</span> <span class="n">P</span><span class="o">.</span><span class="n">dataset</span><span class="p">(</span><span class="n">tile_px</span><span class="o">=</span><span class="mi">299</span><span class="p">,</span> <span class="n">tile_um</span><span class="o">=</span><span class="mi">302</span><span class="p">)</span>
<span class="c1"># Generate features using CTransPath</span>
<span class="n">ctranspath</span> <span class="o">=</span> <span class="n">sf</span><span class="o">.</span><span class="n">build_feature_extractor</span><span class="p">(</span><span class="s1">'ctranspath'</span><span class="p">,</span> <span class="n">resize</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="n">features</span> <span class="o">=</span> <span class="n">sf</span><span class="o">.</span><span class="n">DatasetFeatures</span><span class="p">(</span><span class="n">ctranspath</span><span class="p">,</span> <span class="n">dataset</span><span class="o">=</span><span class="n">dataset</span><span class="p">)</span>
<span class="n">features</span><span class="o">.</span><span class="n">to_torch</span><span class="p">(</span><span class="s1">'/path/to/bag_directory'</span><span class="p">)</span>
<span class="c1"># Evaluate a saved MIL model</span>
<span class="n">df</span> <span class="o">=</span> <span class="n">P</span><span class="o">.</span><span class="n">evaluate_mil</span><span class="p">(</span>
<span class="s1">'/path/to/saved_model'</span>
<span class="n">outcomes</span><span class="o">=</span><span class="s1">'HPV_status'</span><span class="p">,</span>
<span class="n">dataset</span><span class="o">=</span><span class="n">dataset</span><span class="p">,</span>
<span class="n">bags</span><span class="o">=</span><span class="s1">'/path/to/bag_directory'</span><span class="p">,</span>
<span class="p">)</span>
</pre></div>
</div>
<p>As with training, attention heatmaps can be generated for attention-based MIL models with the argument <code class="docutils literal notranslate"><span class="pre">attention_heatmaps=True</span></code>, and these can be customized using <code class="docutils literal notranslate"><span class="pre">cmap</span></code> and <code class="docutils literal notranslate"><span class="pre">interpolation</span></code> arguments.</p>
<img alt="../_images/att_heatmap.jpg" src="../_images/att_heatmap.jpg" />
</section>
<section id="generating-predictions">
<h2>Generating Predictions<a class="headerlink" href="#generating-predictions" title="Permalink to this heading">¶</a></h2>
<p>In addition to generating slide-level predictions during training and evaluation, you can also generate tile-level predictions and attention scores for a dataset using <a class="reference internal" href="../mil_module/#slideflow.mil.get_mil_tile_predictions" title="slideflow.mil.get_mil_tile_predictions"><code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.mil.get_mil_tile_predictions()</span></code></a>. This function returns a DataFrame containing tile-level predictions and attention.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">slideflow.mil</span> <span class="kn">import</span> <span class="n">get_mil_tile_predictions</span>
<span class="gp">>>> </span><span class="n">df</span> <span class="o">=</span> <span class="n">get_mil_tile_predictions</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">dataset</span><span class="p">,</span> <span class="n">bags</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">df</span>
<span class="go"> slide loc_x loc_y ... y_pred3 y_pred4 y_pred5</span>
<span class="go">0 TCGA-4V-A9QI-01Z-0... 2210 7349 ... 0.181155 0.468446 0.070175</span>
<span class="go">1 TCGA-4V-A9QI-01Z-0... 5795 1971 ... 0.243721 0.131991 0.009169</span>
<span class="go">2 TCGA-4V-A9QI-01Z-0... 6273 5437 ... 0.096196 0.583367 0.090258</span>
<span class="go">3 TCGA-4V-A9QI-01Z-0... 2330 3047 ... 0.056426 0.264386 0.300199</span>
<span class="go">4 TCGA-4V-A9QI-01Z-0... 3644 3525 ... 0.134535 0.534353 0.013619</span>
<span class="go">... ... ... ... ... ... ... ...</span>
<span class="go">391809 TCGA-4X-A9FA-01Z-0... 6034 3352 ... 0.004119 0.003636 0.005673</span>
<span class="go">391810 TCGA-4X-A9FA-01Z-0... 6643 1401 ... 0.012790 0.010269 0.011726</span>
<span class="go">391811 TCGA-4X-A9FA-01Z-0... 5546 2011 ... 0.009777 0.013556 0.025255</span>
<span class="go">391812 TCGA-4X-A9FA-01Z-0... 6277 2864 ... 0.026638 0.018499 0.031061</span>
<span class="go">391813 TCGA-4X-A9FA-01Z-0... 4083 4205 ... 0.009875 0.009582 0.022125</span>
<span class="go">[391814 rows x 15 columns]</span>
</pre></div>
</div>
</section>
<section id="single-slide-inference">
<h2>Single-Slide Inference<a class="headerlink" href="#single-slide-inference" title="Permalink to this heading">¶</a></h2>
<p>Predictions can also be generated for individual slides, without requiring the user to manually generate feature bags. Use <code class="xref py py-func docutils literal notranslate"><span class="pre">slideflow.model.predict_slide()</span></code> to generate predictions for a single slide. The first argument is th path to the saved MIL model (a directory containing <code class="docutils literal notranslate"><span class="pre">mil_params.json</span></code>), and the second argument can either be a path to a slide or a loaded <code class="xref py py-class docutils literal notranslate"><span class="pre">sf.WSI</span></code> object.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">slideflow.mil</span> <span class="kn">import</span> <span class="n">predict_slide</span>
<span class="kn">from</span> <span class="nn">slideflow.slide</span> <span class="kn">import</span> <span class="n">qc</span>
<span class="c1"># Load a slide and apply Otsu thresholding</span>
<span class="n">slide</span> <span class="o">=</span> <span class="s1">'/path/to/slide.svs'</span>
<span class="n">wsi</span> <span class="o">=</span> <span class="n">sf</span><span class="o">.</span><span class="n">WSI</span><span class="p">(</span><span class="n">slide</span><span class="p">,</span> <span class="n">tile_px</span><span class="o">=</span><span class="mi">299</span><span class="p">,</span> <span class="n">tile_um</span><span class="o">=</span><span class="mi">302</span><span class="p">)</span>
<span class="n">wsi</span><span class="o">.</span><span class="n">qc</span><span class="p">(</span><span class="n">qc</span><span class="o">.</span><span class="n">Otsu</span><span class="p">())</span>
<span class="c1"># Calculate predictions and attention heatmap</span>
<span class="n">model</span> <span class="o">=</span> <span class="s1">'/path/to/mil_model'</span>
<span class="n">y_pred</span><span class="p">,</span> <span class="n">y_att</span> <span class="o">=</span> <span class="n">predict_slide</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">wsi</span><span class="p">)</span>
</pre></div>
</div>
<p>The function will return a tuple of predictions and attention heatmaps. If the model is not attention-based, the attention heatmap will be <code class="docutils literal notranslate"><span class="pre">None</span></code>. To calculate attention for a model, set <code class="docutils literal notranslate"><span class="pre">attention=True</span></code>:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">y_pred</span><span class="p">,</span> <span class="n">y_att</span> <span class="o">=</span> <span class="n">predict_slide</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">slide</span><span class="p">,</span> <span class="n">attention</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
</pre></div>
</div>
<p>The returned attention values will be a masked <code class="docutils literal notranslate"><span class="pre">numpy.ndarray</span></code> with the same shape as the slide tile extraction grid. Unused tiles will have masked attention values.</p>
</section>
<section id="visualizing-predictions">
<h2>Visualizing Predictions<a class="headerlink" href="#visualizing-predictions" title="Permalink to this heading">¶</a></h2>
<p>Heatmaps of attention and tile-level predictions can be interactively visualized in Slideflow Studio by enabling the Multiple-Instance Learning extension (new in Slideflow 2.1.0). This extension is discussed in more detail in the <a class="reference internal" href="../studio/#extensions"><span class="std std-ref">Extensions</span></a> section.</p>
</section>
</section>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="../ssl/" class="btn btn-neutral float-right" title="Self-Supervised Learning (SSL)" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="../features/" class="btn btn-neutral" title="Generating Features" accesskey="p" rel="prev"><img src="../_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<hr>
<div role="contentinfo">
<p>
© Copyright 2023, James M Dolezal.
</p>
</div>
<div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</div>
</footer>
</div>
</div>
<div class="pytorch-content-right" id="pytorch-content-right">
<div class="pytorch-right-menu" id="pytorch-right-menu">
<div class="pytorch-side-scroll" id="pytorch-side-scroll-right">
<ul>
<li><a class="reference internal" href="#">Multiple-Instance Learning (MIL)</a><ul>
<li><a class="reference internal" href="#generating-features">Generating Features</a></li>
<li><a class="reference internal" href="#training">Training</a><ul>
<li><a class="reference internal" href="#model-configuration">Model Configuration</a></li>
<li><a class="reference internal" href="#classification-regression">Classification & Regression</a></li>
<li><a class="reference internal" href="#training-an-mil-model">Training an MIL Model</a></li>
<li><a class="reference internal" href="#multi-magnification-mil">Multi-Magnification MIL</a></li>
<li><a class="reference internal" href="#custom-architectures">Custom Architectures</a></li>
</ul>
</li>
<li><a class="reference internal" href="#evaluation">Evaluation</a></li>
<li><a class="reference internal" href="#generating-predictions">Generating Predictions</a></li>
<li><a class="reference internal" href="#single-slide-inference">Single-Slide Inference</a></li>
<li><a class="reference internal" href="#visualizing-predictions">Visualizing Predictions</a></li>
</ul>
</li>
</ul>
</div>
</div>
</div>
</section>
</div>
<script type="text/javascript" id="documentation_options" data-url_root="../" src="../_static/documentation_options.js"></script>
<script data-url_root="../" id="documentation_options" src="../_static/documentation_options.js"></script>
<script src="../_static/doctools.js"></script>
<script src="../_static/sphinx_highlight.js"></script>
<script type="text/javascript" src="../_static/js/vendor/jquery-3.6.3.min.js"></script>
<script type="text/javascript" src="../_static/js/vendor/popper.min.js"></script>
<script type="text/javascript" src="../_static/js/vendor/bootstrap.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/list.js/1.5.0/list.min.js"></script>
<script type="text/javascript" src="../_static/js/theme.js"></script>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
<!-- Begin Footer -->
<!-- End Footer -->
<!-- Begin Mobile Menu -->
<div class="mobile-main-menu">
<div class="container-fluid">
<div class="container">
<div class="mobile-main-menu-header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<a class="main-menu-close-button" href="#" data-behavior="close-mobile-menu"></a>
</div>
</div>
</div>
<div class="mobile-main-menu-links-container">
<div class="main-menu">
<ul>
<li>
<a href="https://slideflow.dev">Docs</a>
</li>
<li>
<a href="https://slideflow.dev/tutorial1/">Tutorials</a>
</li>
<li>
<a href="https://github.com/slideflow/slideflow">Github</a>
</li>
</ul>
</div>
</div>
</div>
<!-- End Mobile Menu -->
<script script type="text/javascript">
var collapsedSections = [];
</script>
<script type="text/javascript" src="../_static/js/vendor/anchor.min.js"></script>
<script type="text/javascript">
$(document).ready(function() {
mobileMenu.bind();
mobileTOC.bind();
pytorchAnchors.bind();
sideMenus.bind();
scrollToAnchor.bind();
highlightNavigation.bind();
mainMenuDropdown.bind();
filterTags.bind();
// Add class to links that have code blocks, since we cannot create links in code blocks
$("article.pytorch-article a span.pre").each(function(e) {
$(this).closest("a").addClass("has-code");
});
})
</script>
</body>
</html>
Datasets
Models
