Algorithm 1. Pseudocode for adaptive data structure in RL agents.

1: Input: Full dataset D, window configurations $windows=\{\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{l}\mathrm{y},\mathrm{s}\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{n}\mathrm{u}\mathrm{a}\mathrm{l},\mathrm{q}\mathrm{u}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{y}\}$

2: Output: Trained agents for specific time periods

3:

4: Function TrainRLAgnets (D, windows):

5:    $data\mathrm{\_}splits\leftarrow$ SplitDataset(D, windows)

6:    $agents\leftarrow \{\}$ {Initialize empty dictionary for storing agents}

7:   for each $window\mathrm{\_}type$ in windows:

8:     for each segment in $data\mathrm{\_}splits[window\mathrm{\_}type]$:

9:        $agent\leftarrow$ InitializeRLModel()

10:       Train(agent, segment.data) {Train agent on data segment}

11:        $agents[segment]\leftarrow agent$

12:   return agents

13:

14: Function SplitDataset(D, windows):

15:    $data\mathrm{\_}splits\leftarrow \{\}$

16:   for each $window\mathrm{\_}type$ in windows:

17:     if $window\mathrm{\_}type==$ “yearly”:

18:        $data\mathrm{\_}splits[window\mathrm{\_}type]\leftarrow$ SplitByYear(D)

19:     if $window\mathrm{\_}type==$ "semiannual":

20:        $data\mathrm{\_}splits[window\mathrm{\_}type]\leftarrow$ SplitBySixMonths(D)

21:     if $window\mathrm{\_}type==$ "quarterly":

22:        $data\mathrm{\_}splits[window\mathrm{\_}type]\leftarrow$ SplitByThreeMonths(D)

23:   return $data\mathrm{\_}splits$

24:

25: Main Program:

26: $D\leftarrow$ LoadMarketData()

27: $windows\leftarrow \{\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{l}\mathrm{y},\mathrm{s}\mathrm{e}\mathrm{m}\mathrm{i}\mathrm{a}\mathrm{n}\mathrm{n}\mathrm{u}\mathrm{a}\mathrm{l},\mathrm{q}\mathrm{u}\mathrm{a}\mathrm{r}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{y}\}$

28: $trained\mathrm{\_}agents\leftarrow$ TrainRLAgents(D, windows)

29: Print(“Agents trained:", $trained\mathrm{\_}agents.keys()$)