Stadium relaunch: full 127-page deep-dive report -- per-cell, per-ISA,
every factor interaction, and a raw-data appendix Expanded the campaign-mechanism validation report from a condensed 6-page summary into the full depth Captain Bob asked for: analyze all 9 cells as a conglomerate Latin square, then dive into each cell's own data, then cover every within-ISA and cross-factor interaction explicitly rather than averaging it away. Report structure (127 pages, compiled clean, no undefined references): - Front matter: context, methodology, the SWAP-MTX bug narrative (console-interleaving fix + the Fisher-Yates correctness bug and its fix, both already committed separately) - Layer 1: aggregate 3x3 Latin square (heatmap, invariant-metrics table) - Per-Cell Deep Dive (9 sections): cfg-level distribution, summary table, and a rep-order execution-trajectory chart per cell -- the trajectory charts are what actually visualize the order-dependence finding rather than just stating it - Per-ISA Deep Dive (3 sections): within-architecture seed comparison (violin plots, Kruskal-Wallis, per-factor main effects) - Factor Interactions (6 sections, every pairwise combination of the 4 L8 binary factors): both infer_dec_q and early_exit interaction plots faceted by architecture, plus the three-way factor x factor x architecture significance test - Per-Factor Response (4 sections): linear response by architecture, with an explicit note that a true quadratic term isn't identifiable from this 2-level factorial design - Appendix: full run_id-ordered raw data, all 4,320 rows across all 9 cells, as the primary-source backing for every statistic above Generated programmatically (analyse_stadium_relaunch_fixed.R for the aggregate layer, generate_stadium_deepdive.R for the per-cell/per-ISA/ interaction/appendix layers) rather than hand-authored, since content at this scale needs to be data-driven to stay honest. Also includes analyse_stadium_relaunch.R, the earlier script built against the pre-fix (buggy-shuffle) dataset -- superseded but kept for the record, matching how the underlying data commits were handled. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Robert Allan James
co-authored by
Claude Sonnet 5
Claude Sonnet 5
parent
dbe4b671a1
commit
cf5b08fb65
@@ -0,0 +1,9 @@
|
||||
\input{sections/appendix_amd64_seed12345}
|
||||
\input{sections/appendix_amd64_seed67890}
|
||||
\input{sections/appendix_amd64_seed13579}
|
||||
\input{sections/appendix_aarch64_seed12345}
|
||||
\input{sections/appendix_aarch64_seed67890}
|
||||
\input{sections/appendix_aarch64_seed13579}
|
||||
\input{sections/appendix_riscv64_seed12345}
|
||||
\input{sections/appendix_riscv64_seed67890}
|
||||
\input{sections/appendix_riscv64_seed13579}
|
||||
@@ -0,0 +1,9 @@
|
||||
\input{sections/cell_amd64_seed12345}
|
||||
\input{sections/cell_amd64_seed67890}
|
||||
\input{sections/cell_amd64_seed13579}
|
||||
\input{sections/cell_aarch64_seed12345}
|
||||
\input{sections/cell_aarch64_seed67890}
|
||||
\input{sections/cell_aarch64_seed13579}
|
||||
\input{sections/cell_riscv64_seed12345}
|
||||
\input{sections/cell_riscv64_seed67890}
|
||||
\input{sections/cell_riscv64_seed13579}
|
||||
@@ -0,0 +1,6 @@
|
||||
\input{sections/interact_ent_f_cv_f}
|
||||
\input{sections/interact_ent_f_tmp_f}
|
||||
\input{sections/interact_ent_f_stb_f}
|
||||
\input{sections/interact_cv_f_tmp_f}
|
||||
\input{sections/interact_cv_f_stb_f}
|
||||
\input{sections/interact_tmp_f_stb_f}
|
||||
@@ -0,0 +1,3 @@
|
||||
\input{sections/isa_amd64}
|
||||
\input{sections/isa_aarch64}
|
||||
\input{sections/isa_riscv64}
|
||||
@@ -0,0 +1,4 @@
|
||||
\input{sections/quad_ent_in}
|
||||
\input{sections/quad_cv_in}
|
||||
\input{sections/quad_tmp_in}
|
||||
\input{sections/quad_stb_in}
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: aarch64, seed 12345}
|
||||
\label{sec:cell_aarch64_seed12345}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 56.85, sd 29.30, median 68.0 (IQR 56.0--78.0). Early-exit rate: 0.988.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_aarch64_seed12345_detail_light.pdf}
|
||||
\caption{Cell aarch64/seed 12345: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell aarch64/seed 12345.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 50.73 & 29.73 & 0.967 \\
|
||||
1 & 30 & 50.40 & 34.37 & 1.000 \\
|
||||
2 & 30 & 64.87 & 23.55 & 1.000 \\
|
||||
3 & 30 & 61.80 & 26.20 & 1.000 \\
|
||||
4 & 30 & 51.43 & 32.62 & 1.000 \\
|
||||
5 & 30 & 50.93 & 34.62 & 1.000 \\
|
||||
6 & 30 & 53.47 & 31.13 & 1.000 \\
|
||||
7 & 30 & 55.47 & 31.76 & 0.967 \\
|
||||
8 & 30 & 61.07 & 25.89 & 1.000 \\
|
||||
9 & 30 & 56.30 & 30.21 & 1.000 \\
|
||||
10 & 30 & 60.20 & 28.39 & 0.967 \\
|
||||
11 & 30 & 60.57 & 28.79 & 1.000 \\
|
||||
12 & 30 & 50.23 & 32.09 & 1.000 \\
|
||||
13 & 30 & 52.10 & 30.45 & 1.000 \\
|
||||
14 & 30 & 59.70 & 25.47 & 0.933 \\
|
||||
15 & 30 & 70.40 & 15.51 & 0.967 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_aarch64_seed12345_trajectory_light.pdf}
|
||||
\caption{Cell aarch64/seed 12345: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: aarch64, seed 13579}
|
||||
\label{sec:cell_aarch64_seed13579}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 44.50, sd 35.48, median 68.0 (IQR 0.0--74.0). Early-exit rate: 0.992.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_aarch64_seed13579_detail_light.pdf}
|
||||
\caption{Cell aarch64/seed 13579: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell aarch64/seed 13579.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 53.63 & 32.94 & 0.967 \\
|
||||
1 & 30 & 40.90 & 36.44 & 1.000 \\
|
||||
2 & 30 & 33.67 & 36.65 & 0.967 \\
|
||||
3 & 30 & 39.07 & 37.20 & 1.000 \\
|
||||
4 & 30 & 48.50 & 34.94 & 1.000 \\
|
||||
5 & 30 & 43.87 & 36.46 & 0.967 \\
|
||||
6 & 30 & 33.47 & 36.44 & 1.000 \\
|
||||
7 & 30 & 36.60 & 37.25 & 1.000 \\
|
||||
8 & 30 & 43.53 & 36.20 & 1.000 \\
|
||||
9 & 30 & 50.97 & 33.99 & 1.000 \\
|
||||
10 & 30 & 48.90 & 35.20 & 1.000 \\
|
||||
11 & 30 & 48.13 & 34.69 & 1.000 \\
|
||||
12 & 30 & 52.43 & 32.26 & 0.967 \\
|
||||
13 & 30 & 55.93 & 31.45 & 1.000 \\
|
||||
14 & 30 & 43.60 & 36.26 & 1.000 \\
|
||||
15 & 30 & 38.80 & 36.95 & 1.000 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_aarch64_seed13579_trajectory_light.pdf}
|
||||
\caption{Cell aarch64/seed 13579: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: aarch64, seed 67890}
|
||||
\label{sec:cell_aarch64_seed67890}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 28.09, sd 38.34, median 0.0 (IQR 0.0--77.0). Early-exit rate: 0.990.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_aarch64_seed67890_detail_light.pdf}
|
||||
\caption{Cell aarch64/seed 67890: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell aarch64/seed 67890.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 29.33 & 39.24 & 1.000 \\
|
||||
1 & 30 & 21.13 & 35.67 & 1.000 \\
|
||||
2 & 30 & 34.97 & 40.70 & 1.000 \\
|
||||
3 & 30 & 32.03 & 39.93 & 1.000 \\
|
||||
4 & 30 & 32.37 & 40.36 & 0.967 \\
|
||||
5 & 30 & 34.27 & 39.88 & 1.000 \\
|
||||
6 & 30 & 26.83 & 38.61 & 1.000 \\
|
||||
7 & 30 & 40.17 & 40.90 & 1.000 \\
|
||||
8 & 30 & 13.50 & 30.72 & 1.000 \\
|
||||
9 & 30 & 27.00 & 38.86 & 1.000 \\
|
||||
10 & 30 & 26.47 & 38.09 & 1.000 \\
|
||||
11 & 30 & 27.20 & 39.16 & 0.967 \\
|
||||
12 & 30 & 29.80 & 39.86 & 0.967 \\
|
||||
13 & 30 & 29.27 & 39.15 & 1.000 \\
|
||||
14 & 30 & 23.83 & 37.06 & 1.000 \\
|
||||
15 & 30 & 21.20 & 35.77 & 0.933 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_aarch64_seed67890_trajectory_light.pdf}
|
||||
\caption{Cell aarch64/seed 67890: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: amd64, seed 12345}
|
||||
\label{sec:cell_amd64_seed12345}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 56.85, sd 29.30, median 68.0 (IQR 56.0--78.0). Early-exit rate: 0.988.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_amd64_seed12345_detail_light.pdf}
|
||||
\caption{Cell amd64/seed 12345: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell amd64/seed 12345.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 50.73 & 29.73 & 0.967 \\
|
||||
1 & 30 & 50.40 & 34.37 & 1.000 \\
|
||||
2 & 30 & 64.87 & 23.55 & 1.000 \\
|
||||
3 & 30 & 61.80 & 26.20 & 1.000 \\
|
||||
4 & 30 & 51.43 & 32.62 & 1.000 \\
|
||||
5 & 30 & 50.93 & 34.62 & 1.000 \\
|
||||
6 & 30 & 53.47 & 31.13 & 1.000 \\
|
||||
7 & 30 & 55.47 & 31.76 & 0.967 \\
|
||||
8 & 30 & 61.07 & 25.89 & 1.000 \\
|
||||
9 & 30 & 56.30 & 30.21 & 1.000 \\
|
||||
10 & 30 & 60.20 & 28.39 & 0.967 \\
|
||||
11 & 30 & 60.57 & 28.79 & 1.000 \\
|
||||
12 & 30 & 50.23 & 32.09 & 1.000 \\
|
||||
13 & 30 & 52.10 & 30.45 & 1.000 \\
|
||||
14 & 30 & 59.70 & 25.47 & 0.933 \\
|
||||
15 & 30 & 70.40 & 15.51 & 0.967 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_amd64_seed12345_trajectory_light.pdf}
|
||||
\caption{Cell amd64/seed 12345: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: amd64, seed 13579}
|
||||
\label{sec:cell_amd64_seed13579}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 44.50, sd 35.48, median 68.0 (IQR 0.0--74.0). Early-exit rate: 0.992.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_amd64_seed13579_detail_light.pdf}
|
||||
\caption{Cell amd64/seed 13579: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell amd64/seed 13579.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 53.63 & 32.94 & 0.967 \\
|
||||
1 & 30 & 40.90 & 36.44 & 1.000 \\
|
||||
2 & 30 & 33.67 & 36.65 & 0.967 \\
|
||||
3 & 30 & 39.07 & 37.20 & 1.000 \\
|
||||
4 & 30 & 48.50 & 34.94 & 1.000 \\
|
||||
5 & 30 & 43.87 & 36.46 & 0.967 \\
|
||||
6 & 30 & 33.47 & 36.44 & 1.000 \\
|
||||
7 & 30 & 36.60 & 37.25 & 1.000 \\
|
||||
8 & 30 & 43.53 & 36.20 & 1.000 \\
|
||||
9 & 30 & 50.97 & 33.99 & 1.000 \\
|
||||
10 & 30 & 48.90 & 35.20 & 1.000 \\
|
||||
11 & 30 & 48.13 & 34.69 & 1.000 \\
|
||||
12 & 30 & 52.43 & 32.26 & 0.967 \\
|
||||
13 & 30 & 55.93 & 31.45 & 1.000 \\
|
||||
14 & 30 & 43.60 & 36.26 & 1.000 \\
|
||||
15 & 30 & 38.80 & 36.95 & 1.000 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_amd64_seed13579_trajectory_light.pdf}
|
||||
\caption{Cell amd64/seed 13579: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: amd64, seed 67890}
|
||||
\label{sec:cell_amd64_seed67890}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 28.09, sd 38.34, median 0.0 (IQR 0.0--77.0). Early-exit rate: 0.990.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_amd64_seed67890_detail_light.pdf}
|
||||
\caption{Cell amd64/seed 67890: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell amd64/seed 67890.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 29.33 & 39.24 & 1.000 \\
|
||||
1 & 30 & 21.13 & 35.67 & 1.000 \\
|
||||
2 & 30 & 34.97 & 40.70 & 1.000 \\
|
||||
3 & 30 & 32.03 & 39.93 & 1.000 \\
|
||||
4 & 30 & 32.37 & 40.36 & 0.967 \\
|
||||
5 & 30 & 34.27 & 39.88 & 1.000 \\
|
||||
6 & 30 & 26.83 & 38.61 & 1.000 \\
|
||||
7 & 30 & 40.17 & 40.90 & 1.000 \\
|
||||
8 & 30 & 13.50 & 30.72 & 1.000 \\
|
||||
9 & 30 & 27.00 & 38.86 & 1.000 \\
|
||||
10 & 30 & 26.47 & 38.09 & 1.000 \\
|
||||
11 & 30 & 27.20 & 39.16 & 0.967 \\
|
||||
12 & 30 & 29.80 & 39.86 & 0.967 \\
|
||||
13 & 30 & 29.27 & 39.15 & 1.000 \\
|
||||
14 & 30 & 23.83 & 37.06 & 1.000 \\
|
||||
15 & 30 & 21.20 & 35.77 & 0.933 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_amd64_seed67890_trajectory_light.pdf}
|
||||
\caption{Cell amd64/seed 67890: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: riscv64, seed 12345}
|
||||
\label{sec:cell_riscv64_seed12345}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 56.85, sd 29.30, median 68.0 (IQR 56.0--78.0). Early-exit rate: 0.988.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_riscv64_seed12345_detail_light.pdf}
|
||||
\caption{Cell riscv64/seed 12345: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell riscv64/seed 12345.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 50.73 & 29.73 & 0.967 \\
|
||||
1 & 30 & 50.40 & 34.37 & 1.000 \\
|
||||
2 & 30 & 64.87 & 23.55 & 1.000 \\
|
||||
3 & 30 & 61.80 & 26.20 & 1.000 \\
|
||||
4 & 30 & 51.43 & 32.62 & 1.000 \\
|
||||
5 & 30 & 50.93 & 34.62 & 1.000 \\
|
||||
6 & 30 & 53.47 & 31.13 & 1.000 \\
|
||||
7 & 30 & 55.47 & 31.76 & 0.967 \\
|
||||
8 & 30 & 61.07 & 25.89 & 1.000 \\
|
||||
9 & 30 & 56.30 & 30.21 & 1.000 \\
|
||||
10 & 30 & 60.20 & 28.39 & 0.967 \\
|
||||
11 & 30 & 60.57 & 28.79 & 1.000 \\
|
||||
12 & 30 & 50.23 & 32.09 & 1.000 \\
|
||||
13 & 30 & 52.10 & 30.45 & 1.000 \\
|
||||
14 & 30 & 59.70 & 25.47 & 0.933 \\
|
||||
15 & 30 & 70.40 & 15.51 & 0.967 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_riscv64_seed12345_trajectory_light.pdf}
|
||||
\caption{Cell riscv64/seed 12345: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: riscv64, seed 13579}
|
||||
\label{sec:cell_riscv64_seed13579}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 44.50, sd 35.48, median 68.0 (IQR 0.0--74.0). Early-exit rate: 0.992.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_riscv64_seed13579_detail_light.pdf}
|
||||
\caption{Cell riscv64/seed 13579: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell riscv64/seed 13579.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 53.63 & 32.94 & 0.967 \\
|
||||
1 & 30 & 40.90 & 36.44 & 1.000 \\
|
||||
2 & 30 & 33.67 & 36.65 & 0.967 \\
|
||||
3 & 30 & 39.07 & 37.20 & 1.000 \\
|
||||
4 & 30 & 48.50 & 34.94 & 1.000 \\
|
||||
5 & 30 & 43.87 & 36.46 & 0.967 \\
|
||||
6 & 30 & 33.47 & 36.44 & 1.000 \\
|
||||
7 & 30 & 36.60 & 37.25 & 1.000 \\
|
||||
8 & 30 & 43.53 & 36.20 & 1.000 \\
|
||||
9 & 30 & 50.97 & 33.99 & 1.000 \\
|
||||
10 & 30 & 48.90 & 35.20 & 1.000 \\
|
||||
11 & 30 & 48.13 & 34.69 & 1.000 \\
|
||||
12 & 30 & 52.43 & 32.26 & 0.967 \\
|
||||
13 & 30 & 55.93 & 31.45 & 1.000 \\
|
||||
14 & 30 & 43.60 & 36.26 & 1.000 \\
|
||||
15 & 30 & 38.80 & 36.95 & 1.000 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_riscv64_seed13579_trajectory_light.pdf}
|
||||
\caption{Cell riscv64/seed 13579: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,51 @@
|
||||
\subsection{Cell: riscv64, seed 67890}
|
||||
\label{sec:cell_riscv64_seed67890}
|
||||
|
||||
480 rows captured, all 16 \texttt{cfg} values represented exactly 30 times, zero errors. Overall \texttt{infer\_dec\_q}: mean 28.09, sd 38.34, median 0.0 (IQR 0.0--77.0). Early-exit rate: 0.990.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_riscv64_seed67890_detail_light.pdf}
|
||||
\caption{Cell riscv64/seed 67890: \texttt{infer\_dec\_q} distribution (left) and per-config breakdown (right).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-config summary, cell riscv64/seed 67890.}
|
||||
\begin{tabular}{rrrrr}
|
||||
\toprule
|
||||
cfg & n & mean infer\_dec\_q & sd & early-exit rate \\
|
||||
\midrule
|
||||
0 & 30 & 29.33 & 39.24 & 1.000 \\
|
||||
1 & 30 & 21.13 & 35.67 & 1.000 \\
|
||||
2 & 30 & 34.97 & 40.70 & 1.000 \\
|
||||
3 & 30 & 32.03 & 39.93 & 1.000 \\
|
||||
4 & 30 & 32.37 & 40.36 & 0.967 \\
|
||||
5 & 30 & 34.27 & 39.88 & 1.000 \\
|
||||
6 & 30 & 26.83 & 38.61 & 1.000 \\
|
||||
7 & 30 & 40.17 & 40.90 & 1.000 \\
|
||||
8 & 30 & 13.50 & 30.72 & 1.000 \\
|
||||
9 & 30 & 27.00 & 38.86 & 1.000 \\
|
||||
10 & 30 & 26.47 & 38.09 & 1.000 \\
|
||||
11 & 30 & 27.20 & 39.16 & 0.967 \\
|
||||
12 & 30 & 29.80 & 39.86 & 0.967 \\
|
||||
13 & 30 & 29.27 & 39.15 & 1.000 \\
|
||||
14 & 30 & 23.83 & 37.06 & 1.000 \\
|
||||
15 & 30 & 21.20 & 35.77 & 0.933 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\subsubsection{Execution-order trajectory}
|
||||
|
||||
The per-config summary above pools all 30 replications of each config regardless of when they ran. Plotting \texttt{infer\_dec\_q} against \texttt{run\_id} (the sequential execution order through the shuffled 480-row matrix, not the config or replication index) shows whatever path-dependence exists in the underlying rolling-window/decay-slope estimator directly.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{cell_riscv64_seed67890_trajectory_light.pdf}
|
||||
\caption{Cell riscv64/seed 67890: infer\_dec\_q by execution order, coloured by L8 config, with a loess trend.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
\subsection{coefficient of variation $\times$ stability}
|
||||
\label{sec:interact_cv_f_stb_f}
|
||||
|
||||
Three-way interaction term (cv\_f:stb\_f:arch) in the full model: $p=1$ -- no evidence the interaction itself depends on architecture.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_cv_f_stb_f_light.pdf}
|
||||
\caption{coefficient of variation $\times$ stability interaction on mean infer\_dec\_q, faceted by architecture.}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Cell means, coefficient of variation x stability x architecture.}
|
||||
\begin{tabular}{lllrr}
|
||||
\toprule
|
||||
arch & cv\_f & stb\_f & mean & se \\
|
||||
\midrule
|
||||
amd64 & hi & hi & 44.08 & 1.93 \\
|
||||
amd64 & hi & lo & 42.14 & 1.92 \\
|
||||
amd64 & lo & hi & 42.96 & 1.94 \\
|
||||
amd64 & lo & lo & 43.41 & 1.92 \\
|
||||
aarch64 & hi & hi & 44.08 & 1.93 \\
|
||||
aarch64 & hi & lo & 42.14 & 1.92 \\
|
||||
aarch64 & lo & hi & 42.96 & 1.94 \\
|
||||
aarch64 & lo & lo & 43.41 & 1.92 \\
|
||||
riscv64 & hi & hi & 44.08 & 1.93 \\
|
||||
riscv64 & hi & lo & 42.14 & 1.92 \\
|
||||
riscv64 & lo & hi & 42.96 & 1.94 \\
|
||||
riscv64 & lo & lo & 43.41 & 1.92 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_cv_f_stb_f_earlyexit_light.pdf}
|
||||
\caption{coefficient of variation $\times$ stability interaction on \texttt{early\_exit} rate, faceted by architecture -- same factor pair, binary-outcome response.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
\subsection{coefficient of variation $\times$ temporal decay}
|
||||
\label{sec:interact_cv_f_tmp_f}
|
||||
|
||||
Three-way interaction term (cv\_f:tmp\_f:arch) in the full model: $p=1$ -- no evidence the interaction itself depends on architecture.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_cv_f_tmp_f_light.pdf}
|
||||
\caption{coefficient of variation $\times$ temporal decay interaction on mean infer\_dec\_q, faceted by architecture.}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Cell means, coefficient of variation x temporal decay x architecture.}
|
||||
\begin{tabular}{lllrr}
|
||||
\toprule
|
||||
arch & cv\_f & tmp\_f & mean & se \\
|
||||
\midrule
|
||||
amd64 & hi & hi & 41.96 & 1.94 \\
|
||||
amd64 & hi & lo & 44.26 & 1.91 \\
|
||||
amd64 & lo & hi & 44.82 & 1.93 \\
|
||||
amd64 & lo & lo & 41.54 & 1.92 \\
|
||||
aarch64 & hi & hi & 41.96 & 1.94 \\
|
||||
aarch64 & hi & lo & 44.26 & 1.91 \\
|
||||
aarch64 & lo & hi & 44.82 & 1.93 \\
|
||||
aarch64 & lo & lo & 41.54 & 1.92 \\
|
||||
riscv64 & hi & hi & 41.96 & 1.94 \\
|
||||
riscv64 & hi & lo & 44.26 & 1.91 \\
|
||||
riscv64 & lo & hi & 44.82 & 1.93 \\
|
||||
riscv64 & lo & lo & 41.54 & 1.92 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_cv_f_tmp_f_earlyexit_light.pdf}
|
||||
\caption{coefficient of variation $\times$ temporal decay interaction on \texttt{early\_exit} rate, faceted by architecture -- same factor pair, binary-outcome response.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
\subsection{entropy $\times$ coefficient of variation}
|
||||
\label{sec:interact_ent_f_cv_f}
|
||||
|
||||
Three-way interaction term (ent\_f:cv\_f:arch) in the full model: $p=1$ -- no evidence the interaction itself depends on architecture.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_ent_f_cv_f_light.pdf}
|
||||
\caption{entropy $\times$ coefficient of variation interaction on mean infer\_dec\_q, faceted by architecture.}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Cell means, entropy x coefficient of variation x architecture.}
|
||||
\begin{tabular}{lllrr}
|
||||
\toprule
|
||||
arch & ent\_f & cv\_f & mean & se \\
|
||||
\midrule
|
||||
amd64 & hi & hi & 43.94 & 1.90 \\
|
||||
amd64 & hi & lo & 43.65 & 1.93 \\
|
||||
amd64 & lo & hi & 42.28 & 1.95 \\
|
||||
amd64 & lo & lo & 42.71 & 1.93 \\
|
||||
aarch64 & hi & hi & 43.94 & 1.90 \\
|
||||
aarch64 & hi & lo & 43.65 & 1.93 \\
|
||||
aarch64 & lo & hi & 42.28 & 1.95 \\
|
||||
aarch64 & lo & lo & 42.71 & 1.93 \\
|
||||
riscv64 & hi & hi & 43.94 & 1.90 \\
|
||||
riscv64 & hi & lo & 43.65 & 1.93 \\
|
||||
riscv64 & lo & hi & 42.28 & 1.95 \\
|
||||
riscv64 & lo & lo & 42.71 & 1.93 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_ent_f_cv_f_earlyexit_light.pdf}
|
||||
\caption{entropy $\times$ coefficient of variation interaction on \texttt{early\_exit} rate, faceted by architecture -- same factor pair, binary-outcome response.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
\subsection{entropy $\times$ stability}
|
||||
\label{sec:interact_ent_f_stb_f}
|
||||
|
||||
Three-way interaction term (ent\_f:stb\_f:arch) in the full model: $p=1$ -- no evidence the interaction itself depends on architecture.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_ent_f_stb_f_light.pdf}
|
||||
\caption{entropy $\times$ stability interaction on mean infer\_dec\_q, faceted by architecture.}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Cell means, entropy x stability x architecture.}
|
||||
\begin{tabular}{lllrr}
|
||||
\toprule
|
||||
arch & ent\_f & stb\_f & mean & se \\
|
||||
\midrule
|
||||
amd64 & hi & hi & 44.82 & 1.91 \\
|
||||
amd64 & hi & lo & 42.77 & 1.91 \\
|
||||
amd64 & lo & hi & 42.22 & 1.96 \\
|
||||
amd64 & lo & lo & 42.77 & 1.92 \\
|
||||
aarch64 & hi & hi & 44.82 & 1.91 \\
|
||||
aarch64 & hi & lo & 42.77 & 1.91 \\
|
||||
aarch64 & lo & hi & 42.22 & 1.96 \\
|
||||
aarch64 & lo & lo & 42.77 & 1.92 \\
|
||||
riscv64 & hi & hi & 44.82 & 1.91 \\
|
||||
riscv64 & hi & lo & 42.77 & 1.91 \\
|
||||
riscv64 & lo & hi & 42.22 & 1.96 \\
|
||||
riscv64 & lo & lo & 42.77 & 1.92 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_ent_f_stb_f_earlyexit_light.pdf}
|
||||
\caption{entropy $\times$ stability interaction on \texttt{early\_exit} rate, faceted by architecture -- same factor pair, binary-outcome response.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
\subsection{entropy $\times$ temporal decay}
|
||||
\label{sec:interact_ent_f_tmp_f}
|
||||
|
||||
Three-way interaction term (ent\_f:tmp\_f:arch) in the full model: $p=1$ -- no evidence the interaction itself depends on architecture.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_ent_f_tmp_f_light.pdf}
|
||||
\caption{entropy $\times$ temporal decay interaction on mean infer\_dec\_q, faceted by architecture.}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Cell means, entropy x temporal decay x architecture.}
|
||||
\begin{tabular}{lllrr}
|
||||
\toprule
|
||||
arch & ent\_f & tmp\_f & mean & se \\
|
||||
\midrule
|
||||
amd64 & hi & hi & 44.08 & 1.92 \\
|
||||
amd64 & hi & lo & 43.51 & 1.90 \\
|
||||
amd64 & lo & hi & 42.70 & 1.95 \\
|
||||
amd64 & lo & lo & 42.29 & 1.93 \\
|
||||
aarch64 & hi & hi & 44.08 & 1.92 \\
|
||||
aarch64 & hi & lo & 43.51 & 1.90 \\
|
||||
aarch64 & lo & hi & 42.70 & 1.95 \\
|
||||
aarch64 & lo & lo & 42.29 & 1.93 \\
|
||||
riscv64 & hi & hi & 44.08 & 1.92 \\
|
||||
riscv64 & hi & lo & 43.51 & 1.90 \\
|
||||
riscv64 & lo & hi & 42.70 & 1.95 \\
|
||||
riscv64 & lo & lo & 42.29 & 1.93 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_ent_f_tmp_f_earlyexit_light.pdf}
|
||||
\caption{entropy $\times$ temporal decay interaction on \texttt{early\_exit} rate, faceted by architecture -- same factor pair, binary-outcome response.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,43 @@
|
||||
\subsection{temporal decay $\times$ stability}
|
||||
\label{sec:interact_tmp_f_stb_f}
|
||||
|
||||
Three-way interaction term (tmp\_f:stb\_f:arch) in the full model: $p=1$ -- no evidence the interaction itself depends on architecture.
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_tmp_f_stb_f_light.pdf}
|
||||
\caption{temporal decay $\times$ stability interaction on mean infer\_dec\_q, faceted by architecture.}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Cell means, temporal decay x stability x architecture.}
|
||||
\begin{tabular}{lllrr}
|
||||
\toprule
|
||||
arch & tmp\_f & stb\_f & mean & se \\
|
||||
\midrule
|
||||
amd64 & hi & hi & 44.29 & 1.94 \\
|
||||
amd64 & hi & lo & 42.50 & 1.93 \\
|
||||
amd64 & lo & hi & 42.76 & 1.93 \\
|
||||
amd64 & lo & lo & 43.05 & 1.90 \\
|
||||
aarch64 & hi & hi & 44.29 & 1.94 \\
|
||||
aarch64 & hi & lo & 42.50 & 1.93 \\
|
||||
aarch64 & lo & hi & 42.76 & 1.93 \\
|
||||
aarch64 & lo & lo & 43.05 & 1.90 \\
|
||||
riscv64 & hi & hi & 44.29 & 1.94 \\
|
||||
riscv64 & hi & lo & 42.50 & 1.93 \\
|
||||
riscv64 & lo & hi & 42.76 & 1.93 \\
|
||||
riscv64 & lo & lo & 43.05 & 1.90 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=\textwidth]{interact_tmp_f_stb_f_earlyexit_light.pdf}
|
||||
\caption{temporal decay $\times$ stability interaction on \texttt{early\_exit} rate, faceted by architecture -- same factor pair, binary-outcome response.}
|
||||
\end{figure}
|
||||
|
||||
\clearpage
|
||||
|
||||
@@ -0,0 +1,42 @@
|
||||
\subsection{Architecture: aarch64}
|
||||
\label{sec:isa_aarch64}
|
||||
|
||||
All three seeds' cells for aarch64 combined: 1{,}440 rows, zero errors. Seed remains the dominant source of variation within this architecture alone (Kruskal-Wallis $H=50.76$, $p=9.51e-12$) -- identical in shape to the other two architectures (see \S\ref{sec:isa_amd64}--\ref{sec:isa_riscv64}).
|
||||
|
||||
\begin{figure}[h]
|
||||
\centering
|
||||
\includegraphics[width=0.75\textwidth]{isa_aarch64_seed_violin_light.pdf}
|
||||
\caption{aarch64: infer\_dec\_q distribution by seed (violin + inner boxplot).}
|
||||
\end{figure}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-seed summary, aarch64.}
|
||||
\begin{tabular}{lrrrr}
|
||||
\toprule
|
||||
seed & mean & sd & median & early-exit rate \\
|
||||
\midrule
|
||||
12345 & 56.85 & 29.30 & 68.0 & 0.988 \\
|
||||
67890 & 28.09 & 38.34 & 0.0 & 0.990 \\
|
||||
13579 & 44.50 & 35.48 & 68.0 & 0.992 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\begin{table}[h]
|
||||
\centering
|
||||
\caption{Per-factor main-effect means, aarch64 (all seeds pooled).}
|
||||
\begin{tabular}{lrrrr}
|
||||
\toprule
|
||||
factor & mean (hi) & mean (lo) & difference \\
|
||||
\midrule
|
||||
entropy & 43.80 & 42.50 & 1.30 \\
|
||||
coeff.\ of variation & 43.11 & 43.18 & -0.07 \\
|
||||
temporal decay & 43.39 & 42.90 & 0.49 \\
|
||||
stability & 43.52 & 42.77 & 0.75 \\
|
||||
\bottomrule
|
||||
\end{tabular}
|
||||
\end{table}
|
||||
|
||||
\clearpage
|
||||
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user