Data Science and Machine Learning

Field Course (1060) … Why you should (not) take this class

Lukas Schmoigl

lukas.schmoigl@wifo.ac.at

Data Scientist at WIFO

Showcase

What you have to expect

Content

• Coding in R, JavaScript (and Python)

• Topics
  • Coding Setup
  • Databases and APIs
  • Webscraping
  • Data Visualization
  • Supervised Learning and Cross Validation
  • Natural Language Processing

• Why still learn to code in the age of LLMs?

Webscraping

• Extracting data from websites efficiently
• Reading out static pages with Rvest
• Control browser with RSelenium

Data Visualization

dots = FileAttachment("data/ds_data_en.csv").csv({ typed: true})
plotConfig = ({
  height: 520,
  width: 760,
  marginTop: 40,
  marginLeft: 120,
  marginRight: 60,
  fontFamily: "Century Gothic",
  style: {
    fontSize: "15px",
    fontFamily: "Century Gothic",
  },
  x: {
    ticks: 0,
    label: "flexibility →"
  },
  y: {
    ticks: 0,
    label: "↑ ease of use",
  },
  color: {
    type: "identity"
  },
  marks: [
    Plot.ruleX([0]),
    Plot.ruleY([0]),
    Plot.text(
      [
        { x: 0.42, y: 0.42, quadrant: "Do not Use" },
        { x: 0.48, y: 0.48, quadrant: "Use" },
      ],
      {
        x: "x",
        y: "y",
        text: "quadrant",
        textAnchor: "end",
        fill: (d) => (d.quadrant === "Use" ? "#72bb6f" : "red"),
        rotate: 36.6,
        fontSize: 14,
        fontFamily: "Century Gothic",
      }
    ),
        Plot.text(
      [
        { x: 0, y: 0.8, quadrant: "intuitive &\nconcise" },
        { x: 0, y: 0.2, quadrant: "hard to learn &\nverbose" },
        { x: 0.2, y: -0.05, quadrant: "limited" },
        { x: 0.8, y: -0.05, quadrant: "powerful" }
      ],
      {
        x: "x",
        y: "y",
        text: "quadrant",
        dx: -10,
        textAnchor: "end",
        fontSize: 15,
        fontFamily: "Century Gothic"
      }
    ),
    Plot.areaX(
      [
        { x: 0, y: 1 },
        { x: 0.1, y: 0.9 },
        { x: 0.7, y: 0.7 },
        { x: 0.9, y: 0.1 },
        { x: 1, y: 0 }
      ],
      { x: "x", y: "y", curve: "cardinal", fill: "lightgray", fillOpacity: 0.5 }
    ),
    Plot.line(
      [
        { x: 0, y: 0.665 },
        { x: 0.05, y: 0.685 },
        { x: 0.10, y: 0.665 },
        { x: 0.15, y: 0.685 },
        { x: 0.20, y: 0.665 },
        { x: 0.25, y: 0.685 },
        { x: 0.30, y: 0.665 },
        { x: 0.35, y: 0.685 },
        { x: 0.40, y: 0.665 },
        { x: 0.45, y: 0.685 },
        { x: 0.50, y: 0.665 },
        { x: 0.55, y: 0.685 },
        { x: 0.60, y: 0.665 },
        { x: 0.65, y: 0.685 },
        { x: 0.70, y: 0.665 },
        { x: 0.75, y: 0.685 },
        { x: 0.80, y: 0.665 },
        { x: 0.825, y: 0.675 },
      ],
      { 
        x: "x", 
        y: "y",
        curve: "cardinal", 
        strokeDasharray: 4, 
        stroke: "#559bd5",
      }
    ),
    Plot.text(
      [
        { x: 0.84, y: 0.675, quadrant: "↓ Coding" },
      ],
      {
        x: "x",
        y: "y",
        text: "quadrant",
        textAnchor: "start",
        fontSize: 14,
        fontFamily: "Century Gothic",
        fill: "#559bd5",
      }
    ),
    Plot.dot(dots, {
      x: "x",
      y: "y",
      stroke: (d) => (["ggplot2", "vega-lite", "Plot"].includes(d.vis) ? "#72bb6f" : "grey"), 
      href: (d) => `${d.url}`,
      r: 4.5,
    }),
    Plot.text(dots, {
      x: (d) => 
        ["Flourish", "Highcharts"].includes(d.vis) 
          ? d.x - (d.vis.length / 65)
          : ["Excel"].includes(d.vis) 
          ? d.x - (d.vis.length / 65)
          : ["ggplot2"].includes(d.vis) 
          ? d.x - (d.vis.length / 55)
          : d.x,
      y: (d) =>
        d.vis === "ggplot2"
          ? d.y - 0.00
          : d.vis === "Excel"
          ? d.y - 0.05
          : d.vis === "vega-lite"
          ? d.y + 0.00
          : d.y,
      text: "vis",
      dx: 10,
      textAnchor: "start",
      fontSize: 14,
      fontFamily: "Century Gothic",
      fill: (d) => (["ggplot2", "vega-lite", "Plot"].includes(d.vis) ? "#72bb6f" : "grey"),
      href: (d) => `${d.url}`,
    }),
    Plot.rect(
      [
        { x1: 0.49, y1: 0.55, x2: 0.83, y2: 0.65}
      ],{
      x1: "x1", // or ([x1]) => x1
      y1: "y1", // or ([, y1]) => y1
      x2: "x2", // or ([,, x2]) => x2
      y2: "y2", // or ([,,, y2]) => y2
      stroke: "#72bb6f",
      strokeDasharray: 4,
    }),
    Plot.text(
      [
        { x: 0.54, y: 0.52, quadrant: "Grammar of Graphics" },
      ],
      {
        x: "x",
        y: "y",
        text: "quadrant",
        textAnchor: "start",
        fontSize: 14,
        fontFamily: "Century Gothic",
        fill: "#72bb6f",
      }
    ),
    Plot.line(
      [
        { x: 0, y: 0.9 },
        { x: 0.9, y: 0 }
      ],
      {
        x: "x",
        y: "y",
        stroke: "gray",
        strokeDasharray: 4
      }
    ),
    Plot.tip(dots, Plot.pointer({
      x: "x",
      y: "y",
      title: d => `${d.Vorteile}\n―――――――――――――――――――――――\n${d.Nachteile}`, 
      stroke: "transparent",
      fontSize: 14,
      lineWidth: 22
    })),
  ]
})

• Principles of graphical integrity and excellence
• The grammar of graphics

• Examples:
  • FIW Data Dashboard
  • Lebenssituationsindex
  • Revenue Galaxy

Plot.plot(plotConfig)

Machine Learning

complexities = Array.from({ length: 10 }, (_, i) => i + 1);
trainingErrors = complexities.map(c => 2.5 / c + 0.5);
validationErrors = complexities.map(c => 1 + 0.2 + 0.15 * Math.pow(c - 5, 2));

error = ({
  style: { fontSize: "15px", backgroundColor: "transparent", padding: "20px" },
  height: 400,
  width: 500,
  x: {
    label: "Model complexity",
    ticks: [],
    axis: true
  },
  y: {
    label: "Error",
    ticks: [],
    axis: true
  },
  marks: [
    Plot.line(
      complexities.map((c, i) => ({ complexity: c, error: trainingErrors[i] })),
      { x: "complexity", y: "error", stroke: "#9C6B91", curve: "natural" }
    ),
    Plot.line(
      complexities.map((c, i) => ({ complexity: c, error: validationErrors[i] })),
      { x: "complexity", y: "error", stroke: "#336699", curve: "natural" }
    ),
    Plot.text(
      [{ complexity: 8.5, error: 1, text: "Training error" }],
      { x: "complexity", y: "error", text: "text", fill: "#9C6B91", dx: 10, dy: -10 }
    ),
    Plot.text(
      [{ complexity: 8.5, error: 2, text: "Validation error" }],
      { x: "complexity", y: "error", text: "text", fill: "#336699", dx: 10, dy: 10 }
    ),
    Plot.text(
      [{ text: "Underfitting\nHigh bias", complexity: 2.5, error: 4.5 }],
      { x: "complexity", y: "error", text: "text", fill: "black", dx: 5, dy: -10 }
    ),
    Plot.text(
      [{ text: "Overfitting\nHigh variance", complexity: 8.25, error: 4.5 }],
      { x: "complexity", y: "error", text: "text", fill: "black", dx: -5, dy: -10 }
    ),
    Plot.ruleX([1], { stroke: "black" }),
    Plot.ruleY([0])
  ]
});

• Introduction into supervised learning
  • Decision tree learning
  • Bias-variance trade-off
  • Confusion matrix, ROC- and PR-curves
  • Cross validation

• Neural networks
• Natural language processing

• Examples:
  • Image recognition via satellite data

Plot.plot(error)

Formalities

What you need to know

Field Course

• Schedule and resources in index
• Dates and rooms in VVZ

• Grading
  • Assignments (40%)
    • Groups of four
    • In class presentations of solutions (random)
  • Exam (60%)
    • Pen and paper

Seminar

• Kujtim Avdiu Fr 15:00-17:00

• Practical implementations using R
• Group project, presentation and active participation
• Details in VVZ

Prerequisites

• Coding skills in R or a similar language (e.g., Python, Julia)
• Basic understanding of probability, statistics, linear algebra and calculus
• Practical understanding of statistical modeling and experience in working with data