> For the complete documentation index, see [llms.txt](https://ashrithsagar.gitbook.io/classification-fungal10x/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ashrithsagar.gitbook.io/classification-fungal10x/usage/configuration-file.md).

# Configuration file

The configuration file contains various settings and parameters that control the behavior and settings of the project. Refer to the [config-template.yaml](https://github.com/AshrithSagar/Classification-Fungal10x/blob/main/config-template.yaml) file for more information.

## dataset

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>annot_dir</td><td><code>path</code></td><td>Slide annotations directory path. Should have the same names as that in slide_dir.</td></tr><tr><td>create_zip</td><td><code>bool</code></td><td>Bundle the created dataset directory in a ZIP for easier download.</td></tr><tr><td>data_dir_name</td><td><code>str</code></td><td>Used to create <code>dataset/{data_dir_name}/</code>.</td></tr><tr><td>downsample_factor</td><td><code>int</code></td><td>Downsample slides resolution by this factor. Defaults to preserve aspect ratio.</td></tr><tr><td>downsample_size</td><td><code>tuple[int, int]</code></td><td>Downsample slides to this size.</td></tr><tr><td>n_splits</td><td><code>int</code></td><td>Number of splits for cross-validation.</td></tr><tr><td>overlap</td><td><code>bool</code></td><td>Overlap factor for extracting patches. Should be between 0 and 1.</td></tr><tr><td>patch_size</td><td><code>tuple[int, int]</code></td><td>Patch size for the patches.</td></tr><tr><td>save_slides</td><td><code>bool</code></td><td>Whether to save slides, in <code>dataset/{data_dir_name}-slides/</code>.</td></tr><tr><td>slide_dir</td><td><code>path</code></td><td>Slides directory path. Corresponding annotations should be in <code>annot_dir</code>.</td></tr><tr><td>use_augment</td><td><code>bool</code></td><td>Whether to use data augmentation at patch level for the train split. Preferably always use as True.</td></tr></tbody></table>

## gpu

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>device_index</td><td><code>int</code></td><td>Device index for the GPU. Set to -1 to disable GPU and use CPU instead.</td></tr></tbody></table>

## heatmaps

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>alpha</td><td><code>float</code></td><td>Heatmap transparency while overlaying on the slide. Should be between 0 and 1.</td></tr><tr><td>blur</td><td><code>tuple[int, int]</code></td><td>Gaussian blur kernel size for the heatmap.</td></tr><tr><td>cmap</td><td><code>str</code></td><td>Colormap for the heatmap. Refer to matplotlib colormaps.</td></tr><tr><td>downsample_factor</td><td><code>int</code></td><td>Downsample slides resolution by this factor (when <code>source_dir</code> is provided).</td></tr><tr><td>downsample_size</td><td><code>tuple[int, int]</code></td><td>Downsample slides to this size (when <code>source_dir</code> is provided).</td></tr><tr><td>file_extension</td><td><code>str</code></td><td>File extension for the heatmap images to be saved.</td></tr><tr><td>invert_preds</td><td><code>bool</code></td><td>Whether to invert the predictions before making the heatmaps. Default is true.</td></tr><tr><td>overlap</td><td><code>float</code></td><td>Overlap factor for the heatmap patches. Should be between 0 and 1.</td></tr><tr><td>patch_dims</td><td><code>tuple[int, int, int]</code></td><td>Patch dimensions for the heatmap.</td></tr><tr><td>percentile_scale</td><td><code>tuple[int, int]</code></td><td>Scale the heatmap values to percentile using <code>numpy.percentile()</code>.</td></tr><tr><td>percentile_score</td><td><code>bool</code></td><td>Use percentile score for scaling the heatmap values using <code>scipy.stats.percentileofscore()</code>.</td></tr><tr><td>save_dir</td><td><code>path</code></td><td>Directory to save the heatmap images. Will be saved at <code>{exp_base_dir}/{exp_name}/{fold-*}/{save_dir}/</code>.</td></tr><tr><td>source_dir</td><td><code>path</code></td><td>Path to the directory containing the slides. Used to get predictions for the heatmap.</td></tr><tr><td>source_dir_annot</td><td><code>path</code></td><td>Path to the directory containing annotations corresponding to slides in <code>source_dir</code>. Set to null to use slides themselves for heatmaps.</td></tr><tr><td>use_plt</td><td><code>bool</code></td><td>Use matplotlib to generate the heatmap images. If false, heatmaps will match original slide dimensions.</td></tr></tbody></table>

## model

### \_select

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>classifier</td><td><code>str</code></td><td>Model to use for training and inference. Options: {CLAM_SB, EfficientNetB0, MobileNet, ResNet50, VGG16}.</td></tr></tbody></table>

### model-CLAM\_SB

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>k_sample</td><td><code>null</code></td><td></td></tr><tr><td>dropout</td><td><code>null</code></td><td></td></tr><tr><td>learning_rate</td><td><code>null</code></td><td></td></tr><tr><td>loss_weights</td><td><code>dict</code></td><td>Keys: bag, instance</td></tr><tr><td>patience</td><td><code>null</code></td><td></td></tr><tr><td>run_eagerly</td><td><code>null</code></td><td></td></tr></tbody></table>

### model-EfficientNetB0, model-MobileNet, model-ResNet50, model-VGG16

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>freeze_ratio</td><td><code>null</code></td><td></td></tr><tr><td>learning_rate</td><td><code>null</code></td><td></td></tr><tr><td>patience</td><td><code>null</code></td><td></td></tr><tr><td>start_from_epoch</td><td><code>null</code></td><td></td></tr></tbody></table>

## trainer

<table data-full-width="true"><thead><tr><th>Parameter</th><th>Type</th><th>Description</th></tr></thead><tbody><tr><td>batch_size</td><td><code>int</code></td><td>Batch size for training.</td></tr><tr><td>data_dir</td><td><code>path</code></td><td>Path to the directory containing the dataset. Should likely be <code>dataset/{data_dir_name}/</code>.</td></tr><tr><td>evaluate_only</td><td><code>bool</code></td><td>Evaluate the model on the test set only. Useful for evaluating a trained model.</td></tr><tr><td>exp_base_dir</td><td><code>path</code></td><td>Base directory containing all the experiment folders. Usually <code>experiments/</code>.</td></tr><tr><td>exp_name</td><td><code>str</code></td><td>Current experiment name. Will create a directory in <code>exp_base_dir</code> (<code>{exp_base_dir}/{exp_name}/</code>).</td></tr><tr><td>features_dir</td><td><code>path</code></td><td>Path to the directory containing the features, particularly for MIL datasets.</td></tr><tr><td>folds</td><td><code>list[int]</code></td><td>List of folds to be considered. Zero-indexed.</td></tr><tr><td>max_epochs</td><td><code>int</code></td><td>Maximum number of epochs to train the model.</td></tr><tr><td>overwrite_preds</td><td><code>bool</code></td><td>Overwrite predictions if already present in <code>{exp_base_dir}/{exp_name}/{fold-*}/preds.csv</code>.</td></tr><tr><td>patch_dims</td><td><code>tuple[int, int, int]</code></td><td>Patch dimensions of the dataset.</td></tr><tr><td>predictions_file</td><td><code>string</code></td><td>Filename of the predictions CSV file, without the extension.</td></tr><tr><td>save_weights_only</td><td><code>bool</code></td><td>Save only the model's weights during checkpointing. Useful for subclassed models in <code>tf.keras</code>.</td></tr><tr><td>subset_size</td><td><code>int</code></td><td>Subset size of the dataset to use for training. Use <code>null</code> to use the entire dataset.</td></tr><tr><td>use_augment</td><td><code>bool</code></td><td>Whether to use augmented dataset for training (<code>dataset/{data_dir_name}/fold-*/{train}/</code>).</td></tr></tbody></table>