From 04fc4afdbdaec95241779b54aa3ba1f7fa3e5398 Mon Sep 17 00:00:00 2001 From: Niclas Dobbertin Date: Tue, 24 Oct 2023 22:41:55 +0200 Subject: start of experiment analysis --- experiment/analysis/analysis.tex | 148 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 148 insertions(+) create mode 100644 experiment/analysis/analysis.tex (limited to 'experiment/analysis/analysis.tex') diff --git a/experiment/analysis/analysis.tex b/experiment/analysis/analysis.tex new file mode 100644 index 0000000..2896b4f --- /dev/null +++ b/experiment/analysis/analysis.tex @@ -0,0 +1,148 @@ +% Created 2023-10-23 Mon 20:13 +% 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:orgf19bf7c} +\begin{verbatim} +import pandas as pd +import pickle +from pathlib import Path + +\end{verbatim} +\section{Constants} +\label{sec:orgb587203} +\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:org3427b7b} +\begin{verbatim} +def unpickle(pkl): + with open(pkl, "rb") as f: + data = pickle.load(f) + return data +\end{verbatim} +\subsection{Conditions} +\label{sec:org9e15909} +\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:org65d4664} +\begin{verbatim} +data = {} +for condition in conditions: + data[condition] = {} + for vp in (data_path / condition).iterdir(): + data[condition][vp.stem] = unpickle(vp / "vp.pkl") +\end{verbatim} + +\begin{verbatim} +None +\end{verbatim} +\section{Basic statistics} +\label{sec:orgea2a5f1} +\subsection{Total percent correct} +\label{sec:org2eef721} +To find out how well VP solved the tasked, we calculate the accuracy for train +and test phase. + +\begin{verbatim} +def percent_correct(vp): + train = [x for x in vp.keys() if "train" in x] + test = [x for x in vp.keys() if "test" in x] + + train_total = len(train) * len(vp[train[0]]["procedure_order"]) + test_total = len(test) * len(vp[test[0]]["procedure_order"]) + + train_correct = 0 + test_correct = 0 + + def count_correct(trials): + trials_correct = 0 + for sample in trials: + for proc in vp[sample]["procedure_order"]: + vp_ans = vp[sample][proc]["answer"] + for c in vp_ans: + if not c.isdigit(): + vp_ans = vp_ans.replace(c, "") + vp_ans = int(vp_ans) + if vp_ans == vp[sample]["water_sample"][proc][0]: + trials_correct += 1 + return trials_correct + + return count_correct(train) / train_total, count_correct(test) / test_total +\end{verbatim} + +\begin{verbatim} +condition = "random" +df = pd.DataFrame([percent_correct(data[condition][vp]) for vp in data[condition].keys()], columns=["train", "test"]) +df +\end{verbatim} + +\begin{verbatim} + train test +0 0.822222 0.820000 +1 0.966667 0.800000 +2 0.973333 0.980000 +3 0.911111 0.960000 +4 0.906667 0.980000 +5 0.924444 0.943333 +6 0.957778 0.926667 +7 0.857778 0.946667 +8 0.962222 0.970000 +9 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} +pass +\end{verbatim} + +\begin{verbatim} +None +\end{verbatim} +\end{document} \ No newline at end of file -- cgit v1.2.3 From 4c71eec3cd5f5f36c1cdc6d2284f6dd93facc193 Mon Sep 17 00:00:00 2001 From: Niclas Dobbertin Date: Wed, 1 Nov 2023 11:20:45 +0100 Subject: put analysis functions into own py file --- experiment/analysis/analysis.tex | 116 ++++++++++++++++++++------------------- 1 file changed, 59 insertions(+), 57 deletions(-) (limited to 'experiment/analysis/analysis.tex') diff --git a/experiment/analysis/analysis.tex b/experiment/analysis/analysis.tex index 2896b4f..5b56bc8 100644 --- a/experiment/analysis/analysis.tex +++ b/experiment/analysis/analysis.tex @@ -1,4 +1,4 @@ -% Created 2023-10-23 Mon 20:13 +% Created 2023-10-28 Sat 19:43 % Intended LaTeX compiler: pdflatex \documentclass[11pt]{article} \usepackage[utf8]{inputenc} @@ -30,30 +30,26 @@ \tableofcontents \section{Imports} -\label{sec:orgf19bf7c} +\label{sec:orgbdc2c77} \begin{verbatim} import pandas as pd -import pickle from pathlib import Path +from pprint import pprint + +import tools \end{verbatim} \section{Constants} -\label{sec:orgb587203} +\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:org3427b7b} -\begin{verbatim} -def unpickle(pkl): - with open(pkl, "rb") as f: - data = pickle.load(f) - return data -\end{verbatim} +\label{sec:org87e67b0} \subsection{Conditions} -\label{sec:org9e15909} +\label{sec:orga12f2b6} \begin{verbatim} conditions = [x.stem for x in data_path.iterdir() if x.is_dir()] conditions @@ -65,70 +61,55 @@ random & fixed & blocked\\[0pt] \end{tabular} \end{center} \subsection{Data} -\label{sec:org65d4664} +\label{sec:orgcac95cb} \begin{verbatim} data = {} for condition in conditions: data[condition] = {} for vp in (data_path / condition).iterdir(): - data[condition][vp.stem] = unpickle(vp / "vp.pkl") + 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:orgea2a5f1} +\label{sec:org44d0851} \subsection{Total percent correct} -\label{sec:org2eef721} +\label{sec:org461b551} To find out how well VP solved the tasked, we calculate the accuracy for train and test phase. -\begin{verbatim} -def percent_correct(vp): - train = [x for x in vp.keys() if "train" in x] - test = [x for x in vp.keys() if "test" in x] - - train_total = len(train) * len(vp[train[0]]["procedure_order"]) - test_total = len(test) * len(vp[test[0]]["procedure_order"]) - - train_correct = 0 - test_correct = 0 - - def count_correct(trials): - trials_correct = 0 - for sample in trials: - for proc in vp[sample]["procedure_order"]: - vp_ans = vp[sample][proc]["answer"] - for c in vp_ans: - if not c.isdigit(): - vp_ans = vp_ans.replace(c, "") - vp_ans = int(vp_ans) - if vp_ans == vp[sample]["water_sample"][proc][0]: - trials_correct += 1 - return trials_correct - - return count_correct(train) / train_total, count_correct(test) / test_total -\end{verbatim} - \begin{verbatim} condition = "random" -df = pd.DataFrame([percent_correct(data[condition][vp]) for vp in data[condition].keys()], columns=["train", "test"]) +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 -0 0.822222 0.820000 -1 0.966667 0.800000 -2 0.973333 0.980000 -3 0.911111 0.960000 -4 0.906667 0.980000 -5 0.924444 0.943333 -6 0.957778 0.926667 -7 0.857778 0.946667 -8 0.962222 0.970000 -9 0.982222 0.986667 + 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. @@ -139,10 +120,31 @@ present in both, or only one of the two phases. To investigate, we look at the per procedure accuracy per subject. \begin{verbatim} -pass +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} -None + 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 -- cgit v1.2.3