From 0370c6c1e2e41d4c1a86f495daecf90015698eab Mon Sep 17 00:00:00 2001 From: Tim UH Baumeister <67753016+tim-b90@users.noreply.github.com> Date: Sun, 17 Dec 2023 22:05:01 -0800 Subject: [PATCH] Update PLS-Regression.md -correction PLS-R MC description --- mvpaShiny/PLS-Regression.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/mvpaShiny/PLS-Regression.md b/mvpaShiny/PLS-Regression.md index aa20d59..7f3731e 100644 --- a/mvpaShiny/PLS-Regression.md +++ b/mvpaShiny/PLS-Regression.md @@ -45,7 +45,7 @@ The classic **cost function** view shows the **cost function** value on the y-ax ![plsr_cost_function_fraction_distribution](assets/images/PLS-Regression/plsr_cost_function_fraction_distribution.png) -The cost function value distribution plot was developed for *mvpaShiny* to give a deeper insight into the optimal model selection. The x-axis shows comparisons of performance of models having different (increasing) numbers of components, where a model with a higher number of components is compared with a model with a lower number of components (for example 3 versus 2). Each dot shows the prediction error (root mean squared error of prediction = RMSEP) of one model out of the total **number of repetitions** set. The horizontal black lines show the median prediction error for a model with a lower number of components (i.e., n – 1). If a model with a higher number of components (e.g., 3 components) performs better than the model with fewer components (e.g., 2 components), a higher fraction of values are below the horizontal bar. If a model with a higher number of components performs worse than a model with fewer components, a higher fraction of values are above the horizontal bar. Orange dots show the number of better models, whereas blue dots show the number of worse models. The proportion of blue to red dots results in the fraction value, shown in the parentheses. If the value is lower than the selected **validation threshold,** the model with the higher number of components is accepted (i.e., this model have a lower median prediction error than a model with fewer components). However, as stated in the Monte Carlo parameter section above, the starting point (eg 3 vs 4 components) for these comparisons is the overall lowest median **cost function** (default RMSEP) value. +The cost function value distribution plot was developed for *mvpaShiny* to give a deeper insight into the optimal model selection. The x-axis shows comparisons of performance of models having different (increasing) numbers of components, where a model with a higher number of components is compared with a model with a lower number of components (for example 3 versus 2). Each dot shows the prediction error (root mean squared error of prediction = RMSEP) of one model out of the total **number of repetitions** set. The horizontal black lines show the median prediction error for a model with a lower number of components (i.e., n – 1). If a model with a higher number of components (e.g., 3 components) performs better than the model with fewer components (e.g., 2 components), a higher fraction of values are below the horizontal bar (-> green dots). If a model with a higher number of components performs worse than a model with fewer components, a higher fraction of values are above the horizontal bar (-> red dots). The proportion of red to green dots results in the fraction value, shown in the parentheses. If the value is lower than the selected **validation threshold,** the model with the higher number of components is accepted (i.e., this model have a lower median prediction error than a model with fewer components). However, as stated in the Monte Carlo parameter section above, the starting point (eg 3 vs 4 components) for these comparisons is the overall lowest median **cost function** (default RMSEP) value. ### Algebra behind Target projection