Add alt text to scikit-learn documentation

doc/modules/biclustering.rst

@@ -48,6 +48,8 @@ columns:
    :target: ../auto_examples/bicluster/images/sphx_glr_plot_spectral_coclustering_003.png
    :align: center
    :scale: 50
+   :alt: The graph is a square heat map, 5x5, with axes from 0 to 250.  The darkest 5 
+   squares of heat map run diagonally from top left to bottom right.
 
    An example of biclusters formed by partitioning rows and columns.
 
@@ -60,6 +62,8 @@ small:
    :target: ../auto_examples/bicluster/images/sphx_glr_plot_spectral_biclustering_003.png
    :align: center
    :scale: 50
+   :alt: The graph is a square heat map, 5x5, with axes from 0 to 250. The variance of 
+   the values within each bicluster is small, causing a checkerboard effect.
 
    An example of checkerboard biclusters.
 

doc/modules/calibration.rst

@@ -53,6 +53,10 @@ by showing the number of samples in each predicted probability bin.
 .. figure:: ../auto_examples/calibration/images/sphx_glr_plot_compare_calibration_001.png
    :target: ../auto_examples/calibration/plot_compare_calibration.html
    :align: center
+   :alt: Five plots comparing calibration classifiers. The first plot compares 
+   Logistic, Naive Bayes, SVC, and Random forest classifiers side by side with perfect 
+   calibration. The following four histograms isolate each classifier to describe 
+   their predicted probability on the x-axis and count on the y-axis.
 
 .. currentmodule:: sklearn.linear_model
 

doc/modules/clustering.rst

@@ -272,6 +272,10 @@ small, as shown in the example and cited reference.
    :target: ../auto_examples/cluster/plot_mini_batch_kmeans.html
    :align: center
    :scale: 100
+   :alt: A figure of three panels with scatter plots for KMeans, MiniBatchKMeans,
+        and their difference.  Both methods have very similar results, the left and
+        middle panels show the three identified clusters.  The difference
+        panel highlights the <20 points that differ out of the 3000 points. 
 
 
 .. topic:: Examples:
@@ -310,6 +314,7 @@ is given.
    :target: ../auto_examples/cluster/plot_affinity_propagation.html
    :align: center
    :scale: 50
+   :alt: Three distinct clusters created using affinity propagation.
 
 
 Affinity Propagation can be interesting as it chooses the number of
@@ -461,10 +466,15 @@ computed using a function of a gradient of the image.
 .. |noisy_img| image:: ../auto_examples/cluster/images/sphx_glr_plot_segmentation_toy_001.png
     :target: ../auto_examples/cluster/plot_segmentation_toy.html
     :scale: 50
+    :alt: An image of connected nearly equally sized circles. The circles are in 
+    shades of green, blue, and yellow. The entire image is pixelated, blurry.
 
 .. |segmented_img| image:: ../auto_examples/cluster/images/sphx_glr_plot_segmentation_toy_002.png
     :target: ../auto_examples/cluster/plot_segmentation_toy.html
     :scale: 50
+    :alt: An image of connected nearly equally sized circles. Each circle has a 
+    distinct color: dark green, light green, blue, and yellow. The clarity of the image
+     is much better compared to the previous one.
 
 .. centered:: |noisy_img| |segmented_img|
 
@@ -641,10 +651,20 @@ the roll.
 .. |unstructured| image:: ../auto_examples/cluster/images/sphx_glr_plot_ward_structured_vs_unstructured_001.png
         :target: ../auto_examples/cluster/plot_ward_structured_vs_unstructured.html
         :scale: 49
+        :alt: A 3D scatter plot of 1500 data points. Points are located along a swirl
+             or swiss roll. The six clusters are highlighted in different colours. 
+             The hierarchical clustering is performed without connectivity
+             constraints on the structure and is solely based on distance therefore
+             clusters extend to multiple layers of the roll.
 
 .. |structured| image:: ../auto_examples/cluster/images/sphx_glr_plot_ward_structured_vs_unstructured_002.png
         :target: ../auto_examples/cluster/plot_ward_structured_vs_unstructured.html
         :scale: 49
+        :alt: A 3D scatter plot of 1500 data points. Points are located along a swirl
+              or swiss roll. The six clusters are highlighted in different colours.
+              The hierarchical clustering is performed with connectivity constraints
+              that respect the structure of the roll therefore the clusters form a nice
+              parcellation along the roll.
 
 .. centered:: |unstructured| |structured|
 
@@ -798,6 +818,9 @@ by black points below.
 .. |dbscan_results| image:: ../auto_examples/cluster/images/sphx_glr_plot_dbscan_001.png
         :target: ../auto_examples/cluster/plot_dbscan.html
         :scale: 50
+        :alt: A bubble chart that maps three clusters, with the x-axis from -2.5 to 2 
+        and the y-axis from -2 to 2.5. Outliers are indicated separately and are 
+        evenly spaced around the clusters.
 
 .. centered:: |dbscan_results|
 

doc/modules/covariance.rst

@@ -154,6 +154,9 @@ object to the same sample.
    :target: ../auto_examples/covariance/plot_covariance_estimation.html
    :align: center
    :scale: 65%
+   :alt: A plot comparing shrinkage coefficients by regularization parameter: 
+   shrinkage coefficient on the x-axis and error: negative log-likelihood on test data 
+   on the y-axis.
 
    Bias-variance trade-off when setting the shrinkage: comparing the
    choices of Ledoit-Wolf and OAS estimators
@@ -178,6 +181,12 @@ object to the same sample.
    :target: ../auto_examples/covariance/plot_lw_vs_oas.html
    :align: center
    :scale: 75%
+   :alt: Two figures. One of them plots the Mean Squared Error difference between a 
+   LedoitWolf and an OAS estimator of the covariance with the y-axis from 0 to 60.  
+   The other one compares the Shrinkage covariance estimation between the  LedoitWolf  
+   and OAS, with the  y-axis from 0 to 1. In iboth figures the x-axis is the number of 
+   samples from 5 to 30.
+
 
 
 .. _sparse_inverse_covariance:
@@ -210,6 +219,10 @@ cross-validation to automatically set the ``alpha`` parameter.
    :target: ../auto_examples/covariance/plot_sparse_cov.html
    :align: center
    :scale: 60%
+   :alt: Various matrix using different estimators: Empircal covariance, Empircal 
+   precision, Ledoit-Wolf covariance, Ledoit-Wolf precision, GraphicaLassoCV 
+   covariance, GraphicaLassoCV precision, True covariance and True Precision.
+
 
    *A comparison of maximum likelihood, shrinkage and sparse estimates of
    the covariance and precision matrix in the very small samples
@@ -333,10 +346,14 @@ attributes of a :class:`MinCovDet` robust covariance estimator object.
 .. |robust_vs_emp| image:: ../auto_examples/covariance/images/sphx_glr_plot_robust_vs_empirical_covariance_001.png
    :target: ../auto_examples/covariance/plot_robust_vs_empirical_covariance.html
    :scale: 49%
+   :alt: Two line charts comparing estimation errors that are made when using various 
+   types of location and covariance estimates on contaminated Gaussian distributed 
+   data sets.
 
 .. |mahalanobis| image:: ../auto_examples/covariance/images/sphx_glr_plot_mahalanobis_distances_001.png
    :target: ../auto_examples/covariance/plot_mahalanobis_distances.html
    :scale: 49%
+   :alt: Scatterplot with plotted Mahalanobis distances.