Updated docs with thumbnails

Author: KulikDM

CHANGES.txt

@@ -102,3 +102,4 @@ v<1.0.1>, <08/10/2025> -- Updated tests with random state and better test scores
 v<1.0.2>, <10/10/2025> -- Removed Python 3.9 from testing
 v<1.0.2>, <10/10/2025> -- Python 3.13 support
 v<1.0.2>, <11/30/2025> -- Fixed COMB max_contam empty list issue
+V<1.0.3>, <01/19/2026> -- Updated docs with thumbnails

docs/benchmark.rst

@@ -97,7 +97,7 @@ uncertainty about its mean and is the most robust (best least accurate
 prediction). However, for interpretability and general performance the
 ``MIXMOD, FILTER,`` and ``META`` thresholders are good fits.
 
-.. figure:: figs/Benchmark1.png
+.. thumbnail:: figs/Benchmark1.png
    :alt: Benchmark defaults
 
 ----
@@ -117,7 +117,7 @@ dataset with fewer examples and a greater bias.
     :file: tables/Benchmark2.csv
     :class: sphinx-datatable
 
-.. figure:: figs/Benchmark2.png
+.. thumbnail:: figs/Benchmark2.png
    :alt: Benchmark all
 
 ----
@@ -131,7 +131,7 @@ similar setup is followed as the first benchmark test, however, the
 labels were set using the true contamination applied to the decomposed
 scores as the right-hand component of the MCC deterioration equation.
 
-.. figure:: figs/Multi1.png
+.. thumbnail:: figs/Multi1.png
    :alt: Benchmark multiple
 
 However, to effectively compare whether the multiple outlier detection
@@ -147,7 +147,7 @@ From this, it can be shown that by using a multiple outlier likelihood
 score set it generally performs better than using a single outlier
 likelihood scores set.
 
-.. figure:: figs/Multi2.png
+.. thumbnail:: figs/Multi2.png
    :alt: Benchmark multiple comparison
 
 ----
@@ -201,10 +201,10 @@ methods produced results that were comparable to their inputs.
 | COMB5         | COMB(method='stacked')                |
 +---------------+---------------------------------------+
 
-.. figure:: figs/Comb1.png
+.. thumbnail:: figs/Comb1.png
    :alt: Combination Performance
 
-.. figure:: figs/Comb2.png
+.. thumbnail:: figs/Comb2.png
    :alt: Combination Close Up
 
 ----
@@ -232,7 +232,7 @@ potential to over predict will vary significantly based on the selected
 dataset and outlier detection method, and therefore it is important to
 check the predicted contamination level after thresholding.
 
-.. figure:: figs/Overpred.png
+.. thumbnail:: figs/Overpred.png
    :alt: Over prediction
 
 A second over predictive evaluation can also be done, but now with
@@ -243,7 +243,7 @@ even beyond the best contamination level. However, now some clear well
 performing thresholders can be matched to the previous benchmarking,
 notably ``META`` and ``FILTER``.
 
-.. figure:: figs/Overpred_best.png
+.. thumbnail:: figs/Overpred_best.png
    :alt: Over prediction best
 
 ----
@@ -272,7 +272,7 @@ setting different random states (e.g. ``COMB(thresholders =
 DSN(random_state=111222)])``). This should provide a more robust and
 reliable result.
 
-.. figure:: figs/Randomness.png
+.. thumbnail:: figs/Randomness.png
    :alt: Effects of Randomness
 
 ----

docs/conf.py

@@ -64,6 +64,7 @@
     'sphinxcontrib.bibtex',
     'sphinx.ext.napoleon',
     'sphinx_rtd_theme',
+    'sphinxcontrib.images',
     'sphinxcontrib.jquery',
     'sphinx_datatables'
 ]

docs/confidence.rst

@@ -187,12 +187,12 @@ Below are two scatter plots of the results from the example code above.
 However, in the second plot the use of a classification type thresholder
 ``CLF`` has been employed.
 
-.. figure:: figs/Conf1.png
+.. thumbnail:: figs/Conf1.png
    :alt: Scatter plot of the above example 1
 
    Figure 1: Scatter plot of the ``CONF`` evaluated results using ``IQR``.
 
-.. figure:: figs/Conf2.png
+.. thumbnail:: figs/Conf2.png
    :alt: Scatter plot of the above example 2
 
    Figure 2: Scatter plot of the ``CONF`` evaluated results using ``CLF``.

docs/example.rst

@@ -118,7 +118,7 @@ and threshold the outlier detection scores.
              save_figure=False,
          )
 
-.. figure:: figs/KNN_KARCH.png
+.. thumbnail:: figs/KNN_KARCH.png
    :alt: karch demo
 
 ----

docs/index.rst

@@ -220,7 +220,7 @@ Unsupervised Anomaly Detection <https://arxiv.org/abs/2210.10487>`_
 
 **The comparison among of implemented models** is made available below:
 
-.. figure:: figs/All.png
+.. thumbnail:: figs/All.png
    :alt: Comparison of selected models
 
 ############################

docs/ranking.rst

@@ -196,22 +196,22 @@ In order to get a better understanding on how these six proxy-metrics
 performed overall, joyplots below demonstrate the distributions of their
 Pearson's correlation with respect to the MCC scores for each statistic.
 
-.. figure:: figs/Rank1.png
+.. thumbnail:: figs/Rank1.png
    :alt: Total Average Pearson's score
 
    Figure 1: Total average Pearson's score of selected proxy metrics.
 
-.. figure:: figs/Rank2.png
+.. thumbnail:: figs/Rank2.png
    :alt: Mean Pearson's score Across Datasets
 
    Figure 2: Mean Pearson's score across datasets for selected proxy metrics.
 
-.. figure:: figs/Rank3.png
+.. thumbnail:: figs/Rank3.png
    :alt: Median Pearson's score Across Datasets
 
    Figure 3: Median Pearson's score across datasets for selected proxy metrics.
 
-.. figure:: figs/Rank4.png
+.. thumbnail:: figs/Rank4.png
    :alt: Standard Deviation Pearson's score Across Datasets
 
    Figure 4: Standard deviation Pearson's score across datasets for selected proxy metrics.
@@ -253,12 +253,12 @@ RankDCG results to.
 The joyplots below indicate the performance between the methods with
 respect aggregation across each dataset.
 
-.. figure:: figs/Rank5.png
+.. thumbnail:: figs/Rank5.png
    :alt: RankDCG scores for each test
 
    Figure 5: RankDCG scores for each tested combination.
 
-.. figure:: figs/Rank6.png
+.. thumbnail:: figs/Rank6.png
    :alt: Mean RankDCG scores per dataset
 
    Figure 6: Mean RankDCG scores for each tested combination agregated per dataset.
@@ -270,7 +270,7 @@ not overfitted and is able to generalize, the datasets ``mammography,
 skin``, and ``smtp`` were tested as the model had not been trained on
 them.
 
-.. figure:: figs/Rank7.png
+.. thumbnail:: figs/Rank7.png
    :alt: Mean RankDCG scores per test dataset
 
    Figure 7: Mean RankDCG scores for each tested combination aggregated per test dataset.

docs/requirements.txt

@@ -12,6 +12,7 @@ scipy
 sphinx-datatables
 sphinx-rtd-theme
 sphinxcontrib-bibtex
+sphinxcontrib-images
 torch
 tqdm
 xgboost