added rhub workflow

Author: Jouni Helske

.github/workflows/rhub.yaml

@@ -0,0 +1,95 @@
+# R-hub's generic GitHub Actions workflow file. It's canonical location is at
+# https://github.com/r-hub/actions/blob/v1/workflows/rhub.yaml
+# You can update this file to a newer version using the rhub2 package:
+#
+# rhub::rhub_setup()
+#
+# It is unlikely that you need to modify this file manually.
+
+name: R-hub
+run-name: "${{ github.event.inputs.id }}: ${{ github.event.inputs.name || format('Manually run by {0}', github.triggering_actor) }}"
+
+on:
+  workflow_dispatch:
+    inputs:
+      config:
+        description: 'A comma separated list of R-hub platforms to use.'
+        type: string
+        default: 'linux,windows,macos'
+      name:
+        description: 'Run name. You can leave this empty now.'
+        type: string
+      id:
+        description: 'Unique ID. You can leave this empty now.'
+        type: string
+
+jobs:
+
+  setup:
+    runs-on: ubuntu-latest
+    outputs:
+      containers: ${{ steps.rhub-setup.outputs.containers }}
+      platforms: ${{ steps.rhub-setup.outputs.platforms }}
+
+    steps:
+    # NO NEED TO CHECKOUT HERE
+    - uses: r-hub/actions/setup@v1
+      with:
+        config: ${{ github.event.inputs.config }}
+      id: rhub-setup
+
+  linux-containers:
+    needs: setup
+    if: ${{ needs.setup.outputs.containers != '[]' }}
+    runs-on: ubuntu-latest
+    name: ${{ matrix.config.label }}
+    strategy:
+      fail-fast: false
+      matrix:
+        config: ${{ fromJson(needs.setup.outputs.containers) }}
+    container:
+      image: ${{ matrix.config.container }}
+
+    steps:
+      - uses: r-hub/actions/checkout@v1
+      - uses: r-hub/actions/platform-info@v1
+        with:
+          token: ${{ secrets.RHUB_TOKEN }}
+          job-config: ${{ matrix.config.job-config }}
+      - uses: r-hub/actions/setup-deps@v1
+        with:
+          token: ${{ secrets.RHUB_TOKEN }}
+          job-config: ${{ matrix.config.job-config }}
+      - uses: r-hub/actions/run-check@v1
+        with:
+          token: ${{ secrets.RHUB_TOKEN }}
+          job-config: ${{ matrix.config.job-config }}
+
+  other-platforms:
+    needs: setup
+    if: ${{ needs.setup.outputs.platforms != '[]' }}
+    runs-on: ${{ matrix.config.os }}
+    name: ${{ matrix.config.label }}
+    strategy:
+      fail-fast: false
+      matrix:
+        config: ${{ fromJson(needs.setup.outputs.platforms) }}
+
+    steps:
+      - uses: r-hub/actions/checkout@v1
+      - uses: r-hub/actions/setup-r@v1
+        with:
+          job-config: ${{ matrix.config.job-config }}
+          token: ${{ secrets.RHUB_TOKEN }}
+      - uses: r-hub/actions/platform-info@v1
+        with:
+          token: ${{ secrets.RHUB_TOKEN }}
+          job-config: ${{ matrix.config.job-config }}
+      - uses: r-hub/actions/setup-deps@v1
+        with:
+          job-config: ${{ matrix.config.job-config }}
+          token: ${{ secrets.RHUB_TOKEN }}
+      - uses: r-hub/actions/run-check@v1
+        with:
+          job-config: ${{ matrix.config.job-config }}
+          token: ${{ secrets.RHUB_TOKEN }}

vignettes/KFAS.Rnw

@@ -794,7 +794,7 @@ print(varcor, digits = 2)
 
 Parameter estimation of a state space model is often a difficult task, as the likelihood surface contains multiple maxima, thus making the optimization problem highly dependent on the initial values. Often the unknown parameters are related to the unobserved latent states, such as the covariance matrix in this example, with little a priori knowledge. Therefore, it is challenging to guess good initial values, especially in more complex settings. Thus, multiple initial value configurations possibly with several different type of optimization routines is recommended before one can be reasonably sure that proper optimum is found. Here we use the covariance matrix of the observed series as initial values for the covariance structures.
 
-Another issue in the case of non-Gaussian models is the fact that the likelihood computation is based on iterative procedure which is stopped using some stopping criteria (such as the relative change of log-likelihood), so the log-likelihood function actually contains some noise. This in turn can affect the gradient computations in methods like BFGS and can in theory give unreliable results. Using derivative free method like Nelder-Mead is therefore sometimes recommended. On the other hand, BFGS is usually much faster than Nelder-Mead, and thus I prefer to try BFGS first at least in preliminary analysis.
+Another issue in the case of non-Gaussian models is the fact that the likelihood computation is based on iterative procedure which is stopped using some stopping criteria (such as the relative change of log-likelihood), so the log-likelihood function actually contains some noise. This in turn can affect the gradient computations in methods like BFGS can in theory give unreliable results. Using derivative free method like Nelder-Mead is therefore sometimes recommended. On the other hand, BFGS is usually much faster than Nelder-Mead, and thus I prefer to try BFGS first at least in preliminary analysis.
 
 Using the function \code{KFS} we can compute the smoothed estimates of states: