diff options
Diffstat (limited to 'experiment/analysis/analysis.tex')
| -rw-r--r-- | experiment/analysis/analysis.tex | 150 |
1 files changed, 150 insertions, 0 deletions
diff --git a/experiment/analysis/analysis.tex b/experiment/analysis/analysis.tex new file mode 100644 index 0000000..5b56bc8 --- /dev/null +++ b/experiment/analysis/analysis.tex @@ -0,0 +1,150 @@ +% Created 2023-10-28 Sat 19:43 +% Intended LaTeX compiler: pdflatex +\documentclass[11pt]{article} +\usepackage[utf8]{inputenc} +\usepackage[T1]{fontenc} +\usepackage{graphicx} +\usepackage{longtable} +\usepackage{wrapfig} +\usepackage{rotating} +\usepackage[normalem]{ulem} +\usepackage{amsmath} +\usepackage{amssymb} +\usepackage{capt-of} +\usepackage{hyperref} +\author{Niclas Dobbertin} +\date{\today} +\title{Analysis} +\hypersetup{ + pdfauthor={Niclas Dobbertin}, + pdftitle={Analysis}, + pdfkeywords={}, + pdfsubject={}, + pdfcreator={Emacs 29.1 (Org mode 9.7)}, + pdflang={English}} +\usepackage{biblatex} +\addbibresource{/home/niclas/bib/references.bib} +\begin{document} + +\maketitle +\tableofcontents + +\section{Imports} +\label{sec:orgbdc2c77} +\begin{verbatim} +import pandas as pd +from pathlib import Path +from pprint import pprint + +import tools + +\end{verbatim} +\section{Constants} +\label{sec:orgcb8c537} +\begin{verbatim} +data_path = Path("/home/niclas/repos/uni/master_thesis/experiment/data") + +procedures = ["1", "2", "3", "4", "5", "6", "overall"] +\end{verbatim} +\section{Import Data} +\label{sec:org87e67b0} +\subsection{Conditions} +\label{sec:orga12f2b6} +\begin{verbatim} +conditions = [x.stem for x in data_path.iterdir() if x.is_dir()] +conditions +\end{verbatim} + +\begin{center} +\begin{tabular}{lll} +random & fixed & blocked\\[0pt] +\end{tabular} +\end{center} +\subsection{Data} +\label{sec:orgcac95cb} +\begin{verbatim} +data = {} +for condition in conditions: + data[condition] = {} + for vp in (data_path / condition).iterdir(): + data[condition][vp.stem] = tools.unpickle(vp / "vp.pkl") +\end{verbatim} + +\begin{verbatim} +None +\end{verbatim} +\subsection{Useful Subdata} +\label{sec:org4384120} +\begin{verbatim} +# data_correct = {conditons[0]: {}, conditons[1]: {}, conditons[2]: {}} +pass +# for condition in conditions: +# data_correct[condition] = None +\end{verbatim} + +\begin{verbatim} +None +\end{verbatim} +\section{Basic statistics} +\label{sec:org44d0851} +\subsection{Total percent correct} +\label{sec:org461b551} +To find out how well VP solved the tasked, we calculate the accuracy for train +and test phase. + +\begin{verbatim} +condition = "random" +df = pd.DataFrame([tools.total_accuracy(data[condition][vp], procedures) for vp in data[condition].keys()], index=data[condition].keys(), columns=["train", "test"]) +df +\end{verbatim} + +\begin{verbatim} + train test +vp12 0.822222 0.820000 +vp19 0.966667 0.800000 +vp15 0.973333 0.980000 +vp17 0.911111 0.960000 +vp20 0.906667 0.980000 +vp10 0.924444 0.943333 +vp16 0.957778 0.926667 +vp13 0.857778 0.946667 +vp18 0.962222 0.970000 +vp14 0.982222 0.986667 +\end{verbatim} + +Most subjects have an accuracy of over 95\% in both training and test phase. +Some however are notably lower, under 90\% in either training or test phase, or +both. +This could be a systematic misunderstanding of specific equations, that are +present in both, or only one of the two phases. +To investigate, we look at the per procedure accuracy per subject. + +\begin{verbatim} +condition = "random" +proc_accs = [ + tools.count_correct(data[condition][vp], data[condition][vp].keys(), procedures) + for vp in data[condition].keys() +] +for vp in proc_accs: + for proc in vp.keys(): + vp[proc] /= len(next(iter(data[condition].values())).keys()) +df = pd.DataFrame(proc_accs, index=data[condition].keys()) +df +\end{verbatim} + +\begin{verbatim} + 1 2 3 4 5 6 overall +vp12 0.992 0.592 0.392 0.976 0.960 1.000 0.016 +vp19 1.000 0.992 0.000 0.576 0.992 0.992 0.848 +vp15 0.992 0.992 0.960 0.392 0.592 1.000 0.928 +vp17 0.392 0.968 0.584 1.000 1.000 0.992 0.648 +vp20 0.992 0.376 0.952 0.976 0.976 0.560 0.784 +vp10 0.968 0.360 0.592 0.984 0.984 0.992 0.712 +vp16 0.976 0.600 0.376 0.976 0.992 1.000 0.752 +vp13 0.384 0.960 0.928 0.560 0.992 0.968 0.568 +vp18 0.976 0.976 0.960 0.392 0.600 0.984 0.904 +vp14 0.992 0.976 0.992 0.976 0.400 0.600 0.968 +\end{verbatim} + +We can see that most vp have around 2 procedures with accuracy of around 50\% +\end{document}
\ No newline at end of file |