Update examples

quantile_forest/tests/examples/plot_quantile_extrapolation.py

@@ -5,7 +5,8 @@
 An example on a toy dataset that demonstrates that the prediction intervals
 produced by a quantile regression forest do not extrapolate outside of the
 bounds of the data in the training set, an important limitation of the
-approach.
+approach. Notice that the extrapolated median and interval values fail to
+accurately predict values outside of those observed in the training set.
 """
 
 import altair as alt
@@ -164,7 +165,7 @@ def plot_extrapolations(df, title="", legend=False, x_domain=None, y_domain=None
             "y_pred_line": {"type": "line", "color": "#006aff", "name": "Predicted Median"},
             "y_pred_area": {"type": "area", "color": "#e0f2ff", "name": "Predicted 95% Interval"},
             "y_extrp_line": {"type": "line", "color": "red", "name": "Extrapolated Median"},
-            "y_extrp_area": {"type": "area", "color": "red", "name": "Extrapolated Interval"},
+            "y_extrp_area": {"type": "area", "color": "red", "name": "Extrapolated 95% Interval"},
         }
         for k, v in data.items():
             blank = alt.Chart(pd.DataFrame({k: [v["name"]]}))

quantile_forest/tests/examples/plot_quantile_interpolation.py

@@ -2,8 +2,13 @@
 Comparing Quantile Interpolation Methods
 ========================================
 
-An example comparison of interpolation methods that can be applied during
-prediction when the desired quantile lies between two data points.
+An example illustration of the interpolation methods that can be applied
+during prediction when the desired quantile lies between two data points. In
+this toy example, the forest estimator creates a single split that separates
+samples 1–3 and samples 4–5, with quantiles calculated separately for these
+two groups based on the actual sample values. The interpolation methods are
+used when a calculated quantile does not precisely correspond to one of the
+actual values.
 """
 
 import altair as alt

quantile_forest/tests/examples/plot_quantile_multioutput.py

@@ -4,7 +4,9 @@
 
 An example on a toy dataset that demonstrates fitting a single quantile
 regressor for multiple target variables. For each target, multiple quantiles
-can be estimated simultaneously.
+can be estimated simultaneously. In this example, the target variable has
+two output values for each sample, with a single regressor used to estimate
+three quantiles (the median and interval) for each target output.
 """
 
 import altair as alt

quantile_forest/tests/examples/plot_quantile_vs_standard.py

@@ -2,9 +2,12 @@
 Quantile Regression Forests vs. Random Forests
 ==============================================
 
-An example comparison between a quantile regression forest and a standard
-random forest regressor on a synthetic, right-skewed dataset. In a right-
-skewed distribution, the mean is to the right of the median.
+An example comparison between the estimates generated by a quantile regression
+forest and a standard random forest regressor on a synthetic, right-skewed
+dataset. In a right-skewed distribution, the mean is to the right of the
+median. As illustrated by a greater overlap in the frequencies of the actual
+and predicted values, the median estimated by a quantile regressor can be a
+more reliable estimator of a skewed distribution than the mean.
 """
 
 import altair as alt

quantile_forest/tests/examples/plot_quantile_weighting.py

@@ -4,8 +4,10 @@
 
 An example comparison of the prediction runtime when using a quantile
 regression forest with weighted and unweighted quantiles to compute the
-predicted output values. A standard random forest regressor is included for
-comparison.
+predicted output values. While weighted and unweighted quantiles produce
+identical outputs, the relative runtime of the methods depends on the number
+of training samples and the total number of leaf samples used to calculate the
+quantiles. A standard random forest regressor is included for comparison.
 """
 
 import time