diff --git a/.gitignore b/.gitignore index 2c35f34..0730a72 100644 --- a/.gitignore +++ b/.gitignore @@ -64,8 +64,6 @@ data/* # Binary files *.pdf -*.jpg -*.png *.zip *.gz *.tar diff --git a/apa-old-doi-prefix.csl b/apa-old-doi-prefix.csl new file mode 100644 index 0000000..02eba17 --- /dev/null +++ b/apa-old-doi-prefix.csl @@ -0,0 +1,687 @@ + + diff --git a/assets/allisonhorst/hiding.png b/assets/allisonhorst/hiding.png new file mode 100644 index 0000000..f07f31f Binary files /dev/null and b/assets/allisonhorst/hiding.png differ diff --git a/assets/allisonhorst/not_normal.png b/assets/allisonhorst/not_normal.png new file mode 100644 index 0000000..0c68ab0 Binary files /dev/null and b/assets/allisonhorst/not_normal.png differ diff --git a/assets/allisonhorst/the_beginning.png b/assets/allisonhorst/the_beginning.png new file mode 100644 index 0000000..9228c8e Binary files /dev/null and b/assets/allisonhorst/the_beginning.png differ diff --git a/assets/logos_combined.png b/assets/logos_combined.png new file mode 100644 index 0000000..a6b9b9b Binary files /dev/null and b/assets/logos_combined.png differ diff --git a/assets/mcrps_learning/profoc_langs.png b/assets/mcrps_learning/profoc_langs.png new file mode 100644 index 0000000..4a9cb98 Binary files /dev/null and b/assets/mcrps_learning/profoc_langs.png differ diff --git a/assets/mcrps_learning/web_pres.png b/assets/mcrps_learning/web_pres.png new file mode 100644 index 0000000..c0089f0 Binary files /dev/null and b/assets/mcrps_learning/web_pres.png differ diff --git a/assets/voldep/frame.png b/assets/voldep/frame.png new file mode 100644 index 0000000..1076109 Binary files /dev/null and b/assets/voldep/frame.png differ diff --git a/assets/web_pres.png b/assets/web_pres.png new file mode 100644 index 0000000..c0089f0 Binary files /dev/null and b/assets/web_pres.png differ diff --git a/hemf_logo.png b/hemf_logo.png new file mode 100644 index 0000000..779fbfe Binary files /dev/null and b/hemf_logo.png differ diff --git a/index.qmd b/index.qmd index 07c4cf4..a68b7fd 100644 --- a/index.qmd +++ b/index.qmd @@ -17,9 +17,13 @@ format: smaller: true fig-format: svg slide-number: true + self-contained-math: true + crossrefs-hover: true execute: daemon: false highlight-style: github +bibliography: assets/library.bib +csl: apa-old-doi-prefix.csl --- ## Outline @@ -284,7 +288,7 @@ Each day, $t = 1, 2, ... T$ - The experts can be institutions, persons, or models - The forecasts can be point-forecasts (i.e., mean or median) or full predictive distributions - We do not need any assumptions concerning the underlying data -- `r Citet(my_bib, "cesa2006prediction")` +- @cesa2006prediction ::: @@ -307,7 +311,7 @@ The cumulative regret: - Indicates the predictive accuracy of the expert $k$ until time $t$. - Measures how much the forecaster *regrets* not having followed the expert's advice -Popular loss functions for point forecasting `r Citet(my_bib, "gneiting2011making")`: +Popular loss functions for point forecasting @gneiting2011making: :::: {.columns} @@ -366,7 +370,7 @@ with $q\geq 2$ and $x_{+}$ the (vector) of positive parts of $x$. ## Optimality -In stochastic settings, the cumulative Risk should be analyezed `r Citet(my_bib, "wintenberger2017optimal")`: +In stochastic settings, the cumulative Risk should be analyezed @wintenberger2017optimal: \begin{align} \underbrace{\widetilde{\mathcal{R}}_t = \sum_{i=1}^t \mathbb{E}[\ell(\widetilde{X}_{i},Y_i)|\mathcal{F}_{i-1}]}_{\text{Cumulative Risk of Forecaster}} \qquad\qquad\qquad \text{ and } \qquad\qquad\qquad @@ -411,7 +415,7 @@ The forecaster is asymptotically not worse than the best convex combination $\wi Satisfying the convexity property \eqref{eq_opt_conv} comes at the cost of slower possible convergence. -According to `r Citet(my_bib, "wintenberger2017optimal")`, an algorithm has optimal rates with respect to selection \eqref{eq_opt_select} and convex aggregation \eqref{eq_opt_conv} if +According to @wintenberger2017optimal, an algorithm has optimal rates with respect to selection \eqref{eq_opt_select} and convex aggregation \eqref{eq_opt_conv} if \begin{align} \frac{1}{t}\left(\widetilde{\mathcal{R}}_t - \widehat{\mathcal{R}}_{t,\min} \right) & = @@ -432,11 +436,11 @@ Algorithms can statisfy both \eqref{eq_optp_select} and \eqref{eq_optp_conv} dep ## Optimality -According to `r Citet(my_bib, "cesa2006prediction")` EWA \eqref{eq_ewa_general} satisfies the optimal selection convergence \eqref{eq_optp_select} in a deterministic setting if the: +According to @cesa2006prediction EWA \eqref{eq_ewa_general} satisfies the optimal selection convergence \eqref{eq_optp_select} in a deterministic setting if the: - Loss $\ell$ is exp-concave - Learning-rate $\eta$ is chosen correctly -Those results can be converted to stochastic iid settings `r Citet(my_bib, "kakade2008generalization")` `r Citet(my_bib, "gaillard2014second")`. +Those results can be converted to stochastic iid settings @kakade2008generalization, @gaillard2014second. The optimal convex aggregation convergence \eqref{eq_optp_conv} can be satisfied by applying the kernel-trick. Thereby, the loss is linearized: \begin{align} @@ -465,7 +469,7 @@ We apply Bernstein Online Aggregation (BOA). It lets us weaken the exp-concavity \label{eq_crps} \end{align*} -It's strictly proper `r Citet(my_bib, "gneiting2007strictly")`. +It's strictly proper @gneiting2007strictly. Using the CRPS, we can calculate time-adaptive weight $w_{t,k}$. However, what if the experts' performance is not uniform over all parts of the distribution? @@ -513,7 +517,7 @@ For convex losses, BOAG satisfies that there exist a $C>0$ such that for $x>0$ i 1-e^{x} \label{eq_boa_opt_conv} \end{equation} -`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *convex aggregation* \eqref{eq_optp_conv} `r Citet(my_bib, "wintenberger2017optimal")` . +`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *convex aggregation* \eqref{eq_optp_conv} @wintenberger2017optimal. The same algorithm satisfies that there exist a $C>0$ such that for $x>0$ it holds that \begin{equation} @@ -551,7 +555,7 @@ for all $x_1,x_2 \in \mathbb{R}$ and $t>0$ that \mathbb{E}\left[ \left. \left( \alpha(\ell'(x_1, Y_t)(x_1 - x_2))^{2}\right)^{1/\beta} \right|\mathcal{F}_{t-1}\right] \end{align*} -`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *selection* \eqref{eq_optp_select} `r Citet(my_bib, "gaillard2018efficient")`. +`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *selection* \eqref{eq_optp_select} @gaillard2018efficient. ::: @@ -609,7 +613,7 @@ $\mathcal{Q}_p'' = f.$ Additionally, if $f$ is a continuous Lebesgue-density with $f\geq\gamma>0$ for some constant $\gamma>0$ on its support $\text{spt}(f)$ then is $\mathcal{Q}_p$ is $\gamma$-strongly convex. -Strong convexity with $\beta=1$ implies **A2** `r fontawesome::fa("check", fill ="#ffa600")` `r Citet(my_bib, "gaillard2018efficient")` +Strong convexity with $\beta=1$ implies **A2** `r fontawesome::fa("check", fill ="#ffa600")` @gaillard2018efficient ::: @@ -947,7 +951,7 @@ The simulation using the new DGP carried out for different algorithms (1000 runs R_{t,k} & = R_{t-1,k}(1-\xi) + \ell(\widetilde{F}_{t},Y_i) - \ell(\widehat{F}_{t,k},Y_i) \label{eq_regret_forget} \end{align*} -**Fixed Shares** `r Citet(my_bib, "herbster1998tracking")` +**Fixed Shares** @herbster1998tracking - Adding fixed shares to the weights - Shrinkage towards a constant solution @@ -1457,7 +1461,7 @@ $$\widetilde{X}_{t}=\sum_{k=1}^K w_{t,k}\widehat{X}_{t,k}$$ - The experts can be institutions, persons, or models - The forecasts can be point-forecasts (i.e., mean or median) or full predictive distributions - We do not need any assumptions concerning the underlying data -- `r Citet(my_bib, "cesa2006prediction")` +- @cesa2006prediction ::: @@ -1478,7 +1482,7 @@ The cumulative regret: - Indicates the predictive accuracy of expert $k$ until time $t$. - Measures how much the forecaster *regrets* not having followed the expert's advice -Popular loss functions for point forecasting `r Citet(my_bib, "gneiting2011making")`: +Popular loss functions for point forecasting @gneiting2011making: :::: {.columns} @@ -1517,7 +1521,7 @@ An appropriate loss: \label{eq_crps} \end{align*} -It's strictly proper `r Citet(my_bib, "gneiting2007strictly")`. +It's strictly proper @gneiting2007strictly. Using the CRPS, we can calculate time-adaptive weights $w_{t,k}$. However, what if the experts' performance varies in parts of the distribution? @@ -1550,9 +1554,9 @@ Using the CRPS, we can calculate time-adaptive weights $w_{t,k}$. However, what Convergence rates of BOA are: -`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *selection* `r Citet(my_bib, "gaillard2018efficient")`. +`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *selection* @gaillard2018efficient. -`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *convex aggregation* `r Citet(my_bib, "wintenberger2017optimal")`. +`r fontawesome::fa("arrow-right", fill ="#000000")` Almost optimal w.r.t *convex aggregation* @wintenberger2017optimal. ::: @@ -1657,7 +1661,7 @@ knitr::include_graphics("assets/mcrps_learning/algorithm.svg") #### Data -- Day-Ahead electricity price forecasts from `r Citet(my_bib, "marcjasz2022distributional")` +- Day-Ahead electricity price forecasts from @marcjasz2022distributional - Produced using probabilistic neural networks - 24-dimensional distributional forecasts - Distribution assumptions: JSU and Normal @@ -3119,18 +3123,10 @@ Accounting for heteroscedasticity or stabilizing the variance via log transforma -`r fontawesome::fa("newspaper")` `r Citet(my_bib, "berrisch2023modeling")` +`r fontawesome::fa("newspaper")` @berrisch2023modeling ::: :::: -## References - -::: {.scrollable} - -```{r refs1, echo=FALSE, results="asis"} -PrintBibliography(my_bib, .opts = list(style = "text")) -``` - -:::: \ No newline at end of file +## References \ No newline at end of file