pca on exam grades

exam_points_meanfree_unitvar.csv

@@ -0,0 +1,34 @@
+1.408367168897572741e+00;1.757964546987063059e+00;1.412795836484065815e+00;1.539532365180545703e+00;4.848876374569938141e-01
+1.054976144967554541e+00;1.757964546987063059e+00;4.597192801257678485e-01;1.539532365180545703e+00;1.390721685453574885e+00
+1.231671656932563641e+00;6.645963531292558013e-01;4.597192801257678485e-01;1.699118037180968210e+00;1.571888495052891077e+00
+3.481940971075180302e-01;1.029052417748524961e+00;2.556487704114023707e+00;1.858703709181390717e+00;-4.209464105395876454e-01
+1.761758192827591163e+00;-4.287718407285517341e-01;1.031565213940746961e+00;5.820183331780111047e-01;8.472212566556260871e-01
+3.481940971075180302e-01;1.029052417748524961e+00;2.691039688541081443e-01;-3.754956988245231608e-01;1.209554875854258693e+00
+-5.352834627175273585e-01;-6.431577610928257416e-02;1.222180525212406499e+00;7.416040051784336118e-01;1.028388066254942279e+00
+3.481940971075180302e-01;6.645963531292558013e-01;4.597192801257678485e-01;-5.632435482367830620e-02;8.472212566556260871e-01
+7.015851210375363411e-01;-1.157683969967089999e+00;7.848865758244849555e-02;1.539532365180545703e+00;8.472212566556260871e-01
+7.015851210375363411e-01;8.468243854388903813e-01;4.597192801257678485e-01;-3.754956988245231608e-01;4.848876374569938141e-01
+-1.818924387875090753e-01;-6.431577610928257416e-02;6.503345913974275527e-01;5.820183331780111047e-01;6.660544470563101171e-01
+1.054976144967554541e+00;-1.157683969967089999e+00;-1.121266536892111809e-01;9.011896771788560079e-01;8.472212566556260871e-01
+7.015851210375363411e-01;-6.431577610928257416e-02;6.503345913974275527e-01;-5.632435482367830620e-02;3.037208278576775111e-01
+-5.196926822499936365e-03;1.393508482367794121e+00;8.409499026690872014e-01;-3.754956988245231608e-01;-4.209464105395876454e-01
+-1.818924387875090753e-01;-1.157683969967089999e+00;8.409499026690872014e-01;4.224326611775887086e-01;6.660544470563101171e-01
+1.408367168897572741e+00;6.645963531292558013e-01;-1.446433832590828805e+00;4.224326611775887086e-01;-6.021132201389038929e-01
+1.054976144967554541e+00;6.645963531292558013e-01;-8.745878987758498591e-01;-5.350813708249456679e-01;1.225540182583612220e-01
+1.714985851425088748e-01;-1.157683969967089999e+00;6.503345913974275527e-01;1.032613171767441246e-01;4.848876374569938141e-01
+-1.818924387875090753e-01;-1.157683969967089999e+00;-4.933572762325305061e-01;7.416040051784336118e-01;1.209554875854258693e+00
+-7.119789746825364585e-01;6.645963531292558013e-01;1.222180525212406499e+00;-6.946670428253680640e-01;-7.832800297382201959e-01
+-1.818924387875090753e-01;-1.157683969967089999e+00;7.848865758244849555e-02;-3.754956988245231608e-01;1.209554875854258693e+00
+-3.585879507525182031e-01;6.645963531292558013e-01;-6.839725875041902103e-01;-5.632435482367830620e-02;-2.397796009402713424e-01
+1.714985851425088748e-01;-7.932279053478209496e-01;-4.933572762325305061e-01;5.820183331780111047e-01;-7.832800297382201959e-01
+3.481940971075180302e-01;-1.157683969967089999e+00;-1.255818521319169268e+00;5.820183331780111047e-01;-5.861279134095506022e-02
+7.015851210375363411e-01;-1.157683969967089999e+00;1.031565213940746961e+00;-1.492595402827480155e+00;-7.832800297382201959e-01
+-1.818924387875090753e-01;-6.431577610928257416e-02;-6.839725875041902103e-01;-1.333009730827057870e+00;3.037208278576775111e-01
+-1.948847558437600380e+00;8.468243854388903813e-01;-6.839725875041902103e-01;-3.754956988245231608e-01;-6.021132201389038929e-01
+-3.585879507525182031e-01;-1.157683969967089999e+00;-1.446433832590828805e+00;-3.754956988245231608e-01;-6.021132201389038929e-01
+-1.418761022542573080e+00;-1.157683969967089999e+00;-3.027419649608708574e-01;-5.350813708249456679e-01;-7.832800297382201959e-01
+-2.125543070402609480e+00;1.393508482367794121e+00;-1.065203210047509508e+00;-1.173424058826635363e+00;-1.326780458536169105e+00
+-1.242065510577563980e+00;6.645963531292558013e-01;-1.446433832590828805e+00;-1.333009730827057870e+00;-1.870280887334117903e+00
+-8.886744866475456694e-01;-1.157683969967089999e+00;-8.745878987758498591e-01;-8.542527148257904601e-01;-1.507947268135485297e+00
+-1.772152046472591280e+00;3.001402885099865858e-01;-1.065203210047509508e+00;-1.971352418828747455e+00;-1.870280887334117903e+00
+-1.242065510577563980e+00;-7.932279053478209496e-01;-1.446433832590828805e+00;-1.492595402827480155e+00;-1.870280887334117903e+00

index.ipynb

@@ -70,6 +70,7 @@
     "- [pca_2D.ipynb](pca_2D.ipynb)\n",
     "- [pca_3D.ipynb](pca_3D.ipynb)\n",
     "- [pca_audio_features.ipynb](pca_audio_features.ipynb)\n",
+    "- [pca_of_exam_grades.ipynb](pca_of_exam_grades.ipynb)\n",
     "\n",
     "## Exercise: Bias Variance Trade-Off vs. Model Complexity\n",
     "- [Bias-Variance Trade-Off vs. Model Complexity](bias_variance_linear_regression.ipynb)\n",

pca_of_exam_grades.ipynb

@@ -0,0 +1,309 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Sascha Spors,\n",
+    "Professorship Signal Theory and Digital Signal Processing,\n",
+    "Institute of Communications Engineering (INT),\n",
+    "Faculty of Computer Science and Electrical Engineering (IEF),\n",
+    "University of Rostock,\n",
+    "Germany\n",
+    "\n",
+    "# Data Driven Audio Signal Processing - A Tutorial with Computational Examples\n",
+    "\n",
+    "Master Course #24512\n",
+    "\n",
+    "- lecture: https://github.com/spatialaudio/data-driven-audio-signal-processing-lecture\n",
+    "- tutorial: https://github.com/spatialaudio/data-driven-audio-signal-processing-exercise\n",
+    "\n",
+    "Feel free to contact lecturer [email protected]\n",
+    "\n",
+    "# PCA on Achieved Points of Written Examination "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import numpy as np\n",
+    "import scipy\n",
+    "from scipy.linalg import svd, diagsvd\n",
+    "import matplotlib as mpl\n",
+    "import matplotlib.pyplot as plt\n",
+    "\n",
+    "np.set_printoptions(precision=3, sign=' ', suppress=True)\n",
+    "\n",
+    "print(np.__version__)  # tested with 1.26.4\n",
+    "print(scipy.__version__)  # tested with 1.13.1\n",
+    "print(mpl.__version__)  # tested with 3.9.2"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "X = np.loadtxt(open(\"exam_points_meanfree_unitvar.csv\", \"rb\"), delimiter=\";\", skiprows=0)\n",
+    "N, F = X.shape\n",
+    "print(N, F)  # 34 students, 5 tasks for exam on signals & systems, a typical course in electrical engineering bachelor studies\n",
+    "# columns correspond to theses tasks\n",
+    "task_label = ['Task 1: Convolution', 'Task 2: Fourier', 'Task 3: Sampling', 'Task 4: Laplace Domain', 'Task 5: z-Domain']"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# data in exam_points_meanfree_unitvar.csv is already mean-free and columns have var=1\n",
+    "# so the numbers in X do not represent points or percentage,\n",
+    "# but rather encode the performance of the students per task in a normalised way\n",
+    "# X is however sorted: first row belongs to best grade, last row to worst grade\n",
+    "np.mean(X, axis=0), np.std(X, axis=0, ddof=1), np.var(X, axis=0, ddof=1)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# for completeness of PCA algorithm ->\n",
+    "# make X zscore (altough it is already)\n",
+    "mu = np.mean(X, axis=0)\n",
+    "X = X - mu  # de-mean\n",
+    "sigma = np.sqrt(np.sum(X**2, axis=0) / (N-1))\n",
+    "X = X / sigma  # normalise to std=1\n",
+    "np.mean(X, axis=0), np.std(X, axis=0, ddof=1), np.var(X, axis=0, ddof=1)  # check"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "X  # print mean=0 / var=1 data matrix"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# get SVD / CovMatrix stuff\n",
+    "[U, s, Vh] = svd(X)\n",
+    "V = Vh.T  # we don't use Vh later on!\n",
+    "S = diagsvd(s, N, F)  # sing vals matrix\n",
+    "D, _ = np.linalg.eig(X.T @ X / (N-1))  # eig vals\n",
+    "D = -np.sort(-D)  # sort them, then ==\n",
+    "d = s**2 / (N-1)\n",
+    "print(np.allclose(d, D))  # so we go for d later on\n",
+    "\n",
+    "# switch polarities for nicer interpretation ot the\n",
+    "# exam data\n",
+    "V[:,0] *= -1\n",
+    "U[:,0] *= -1\n",
+    "\n",
+    "V[:,2] *= -1\n",
+    "U[:,2] *= -1\n",
+    "\n",
+    "V[:,3] *= -1\n",
+    "U[:,3] *= -1"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# PCA\n",
+    "US = U @ S\n",
+    "PC_Features = US @ diagsvd(1 / np.sqrt(d), F, F)  # normalised such that columns have var 1, aka (normalised) PC scores\n",
+    "print(np.var(PC_Features, axis=0, ddof=1))\n",
+    "#PC_Loadings = (diagsvd(np.sqrt(d), F, F) @ V.T).T  # ==\n",
+    "PC_Loadings = V @ diagsvd(np.sqrt(d), F, F)  # aka PC coeff, not unit-length anymore, but normalised such that it shows correlation between PC_Features and X"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "np.allclose(X, PC_Features @ PC_Loadings.T)  # check correct matrix factorisation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# project an x column vector to a pc feature column -> do this for all options -> get all weights for linear comb of pc features\n",
+    "# correlation uses unit-length vectors\n",
+    "PC_Loadings_manual = np.zeros((F, F))\n",
+    "for row in range(F):\n",
+    "    tmp_x = X[:, row] / np.linalg.norm(X[:, row])\n",
+    "    for column in range(F):\n",
+    "        tmp_pc = PC_Features[:, column] / np.linalg.norm(PC_Features[:, column])\n",
+    "        PC_Loadings_manual[row, column] = np.inner(tmp_pc, tmp_x)\n",
+    "np.allclose(PC_Loadings_manual, PC_Loadings)  # we get the PC_Loadings matrix"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# explained variance\n",
+    "d, np.var(US, axis=0, ddof=1)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# explained cum variance in %\n",
+    "cum_var = np.cumsum(d) / np.sum(d) * 100\n",
+    "cum_var"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Check via Plots"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "plt.figure(figsize=(12,8))\n",
+    "\n",
+    "plt.subplot(2,1,1)\n",
+    "for f in range(F):\n",
+    "    plt.plot(X[:, f], 'o-', color='C'+str(f), label='Task '+str(f+1), ms=3)\n",
+    "plt.legend(loc='lower left')\n",
+    "plt.xticks([0, N-1], labels=['best grade', 'worst grade'])\n",
+    "plt.ylabel('normalised points (mean-free, var=1)')\n",
+    "plt.grid(True)\n",
+    "plt.title(task_label)\n",
+    "\n",
+    "plt.subplot(2,1,2)\n",
+    "for f in range(F):\n",
+    "    plt.plot(US[:, f], 'o-', color='C'+str(f), label='PCA v '+str(f+1), lw=(F-f)*2/3, ms=(F-f)*3/2)\n",
+    "plt.legend(loc='lower left')\n",
+    "plt.xticks([0, N-1], labels=['best grade', 'worst grade'])\n",
+    "plt.ylabel('PC features (mean-free, sorted var)')\n",
+    "plt.xlabel('student index (sorted grade)')\n",
+    "plt.grid(True)\n",
+    "plt.title(['cum var in %:', cum_var])\n",
+    "plt.tight_layout()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# correlation between task and pc\n",
+    "pc_label = ['PC 1', 'PC 2', 'PC 3', 'PC 4', 'PC 5']\n",
+    "cmap = plt.get_cmap('Spectral_r', 8)\n",
+    "fig = plt.figure(figsize=(6,4))\n",
+    "ax = fig.add_subplot(111)\n",
+    "cax = ax.matshow(PC_Loadings, cmap=cmap, vmin=-1, vmax=+1)\n",
+    "fig.colorbar(cax)\n",
+    "ax.set_xticks(np.arange(len(pc_label)))\n",
+    "ax.set_yticks(np.arange(len(task_label)))\n",
+    "ax.set_xticklabels(pc_label)\n",
+    "ax.set_yticklabels(task_label)\n",
+    "ax.set_title('Loading Matrix = PC x contributes to Task y')\n",
+    "plt.tight_layout()\n",
+    "\n",
+    "# a rank 3 approximation of the data,\n",
+    "# i.e. using only PC1, PC2 and PC3 in the linear combination to reconstruct X\n",
+    "# would only change one grade by a 1/3 grade step\n",
+    "# so 85.899 % explained variance would be enough to figure the actual grading\n",
+    "\n",
+    "# PC1 and PC2 might allow an intuitive interpretation:\n",
+    "# students are very well prepared to convolution, Laplace and z-Domain tasks\n",
+    "# as theses tasks are always very similar and definitiely will be queried in the exam\n",
+    "# so, PC1 indidcates the performance on 'fulfilled' expectations and is\n",
+    "# highly correlated with the achieved grade\n",
+    "# the Fourier task and the sampling task were chosen out of a wide range of options\n",
+    "# here students have rather 'unknown' expectations, which is why we need PC 2 to cover this\n",
+    "#\n",
+    "# PC 3 to 5 show positive vs. negative correlations, i.e. mostly one good task vs. one bad task performance\n",
+    "# some of these results are intuitive: we know that students sometimes have preferences for Laplace vs. z-Domain  "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "np.sum(PC_Loadings**2, axis=0)  # that's again the explained variance of the PCs"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "np.sum(PC_Loadings**2, axis=1)  # communalities, must sum to 1 in our normalised handling"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Copyright\n",
+    "\n",
+    "- the notebooks are provided as [Open Educational Resources](https://en.wikipedia.org/wiki/Open_educational_resources)\n",
+    "- feel free to use the notebooks for your own purposes\n",
+    "- the text is licensed under [Creative Commons Attribution 4.0](https://creativecommons.org/licenses/by/4.0/)\n",
+    "- the code of the IPython examples is licensed under the [MIT license](https://opensource.org/licenses/MIT)\n",
+    "- please attribute the work as follows: *Frank Schultz, Data Driven Audio Signal Processing - A Tutorial Featuring Computational Examples, University of Rostock* ideally with relevant file(s), github URL https://github.com/spatialaudio/data-driven-audio-signal-processing-exercise, commit number and/or version tag, year."
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "myddasp",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}