GCI World 2026 September Session4 During Lecture

Summary

The lecture introduces exploratory data analysis (EDA) through data visualization, using Python’s Matplotlib and Seaborn libraries and a city bike-rental dataset. Its central lesson is that visualizations help analysts find patterns and answer practical questions—but a graph is useful only when it is clear, appropriate to the data, and guided by a specific question.

1. EDA and the purpose of visualization

• EDA is an iterative process of analysis and visualization used to understand a dataset, notice patterns or biases, and form and test hypotheses.
• During EDA, speed and flexibility matter more than making every chart presentation-ready. Analysts should try different views, learn from them, and adjust their questions.
• EDA can inform communication, but exploring data and creating polished charts for an audience are related yet distinct activities.
• A useful chart should answer a clear question, be understandable without extra explanation, avoid unnecessary elements, represent the data accurately, and draw attention to what matters.

2. Dataset and initial inspection

The lecture uses daily data from the UCI bike-rental dataset, containing 731 days—about two years—of records.

The initial workflow is to:

• Import NumPy, Pandas, Matplotlib’s pyplot (usually as plt), and Seaborn (usually as sns).
• Load the daily CSV into a Pandas DataFrame.
• Inspect its dimensions and column information, then review summary statistics.
• Treat coded fields such as season, holiday, and working day as categorical variables, even if they are stored as numbers. Their codes represent groups, not quantities for arithmetic.
• Identify cnt as the total daily rental count, the main outcome of interest.

The dataset has no missing values. Daily rental counts average roughly 4,500 but vary widely, which makes visual analysis useful.

3. Matplotlib chart design and structure

A basic line plot can reveal broad patterns, but an unlabeled chart is difficult to interpret. The lecture recommends adding:

• A clear title with plt.title.
• Informative axis labels with plt.xlabel and plt.ylabel.
• An appropriate figure size using figsize.
• A grid when it helps readers estimate values.
• Readable scales and labels, with tick rotation where needed to prevent overlap.

Matplotlib’s layout concepts are also introduced:

• A figure is the overall canvas.
• Axes are the plot areas within the figure.
• Axis objects define the x- and y-scales and their ticks.
• plt.subplot divides a figure into a grid and selects a position. Subplot positions are counted from 1, left to right and then top to bottom.
• plt.tight_layout() helps prevent labels from overlapping in multi-plot figures.
• Non-English characters may not render reliably with some Matplotlib font setups. Using supported fonts or suitable alternatives can avoid display problems.

4. Choosing and interpreting chart types

• Line charts (plt.plot) are useful for time series. Converting the date column to datetime and displaying real dates makes the timeline more meaningful. Monthly tick marks and rotated labels can improve readability.

  • In the bike data, the daily series shows an upward trend across the two years, seasonal peaks in warmer months, and sharp day-to-day fluctuations.
  • Plotting casual and registered users separately shows that registered users account for most rentals and generally have a steadier pattern.
  • Grouping by month or calculating a rolling average, such as a seven-day average, can reduce daily noise and make trends easier to see.

• Histograms (plt.hist) show how frequently values fall into ranges. The bins parameter controls the number of intervals; custom bin boundaries or numpy.arange can create useful ranges. The range parameter can focus the display on values of interest, though excluding outliers may hide information.

  • Bike-rental counts are concentrated mostly around 3,000–6,000 rentals per day; very low and very high counts are less common.
  • Comparing histograms can help assess differences between groups, such as weekdays and weekends.

• Box plots (plt.boxplot) summarize a distribution:

  • The median marks the midpoint.
  • The box spans the 25th to 75th percentiles, containing the middle half of observations.
  • Whiskers show the range of typical values, and points beyond them indicate possible outliers.
  • Horizontal orientation can make some charts easier to read; patch_artist=True can fill the box with color.
  • Comparing weekday and weekend box plots can show differences in typical demand, spread, and unusual days.

• Scatter plots (plt.scatter) display the relationship between two numerical variables. In the dataset, temperature and rentals generally rise together, indicating a positive association, though very hot conditions may not follow the same pattern. Setting alpha makes overlapping points easier to inspect.

• Bubble charts add a third variable by changing point size, such as using wind speed for bubble size. The size scale should be chosen carefully, and transparency can reduce clutter.

• Bar charts (plt.bar and plt.barh) compare numerical summaries across categories. Converting numeric codes to meaningful labels makes them more interpretable. Horizontal bars can help when category labels are long.

  • Grouped bars compare multiple series side by side.
  • Stacked bars use the bottom argument to show both total height and the contribution of each group. The lecture’s weekday/weekend comparison suggests registered users dominate on working days, while casual users make up a larger share on non-working days.

• Pie charts (plt.pie) can emphasize a dominant share, but smaller slices are difficult to compare. Bar charts are often clearer, especially when categories are similar in size.

• Subplots place charts together for direct comparison—for example, histograms of rentals and temperature, or scatter plots of rentals against several weather variables. They should clarify comparisons rather than cram too many small, unreadable charts into one figure.

5. Using Seaborn for grouped and multivariable analysis

Seaborn works with Matplotlib and can simplify statistical plots. Its hue parameter assigns colors to categories and creates a legend, while the data argument allows plots to refer directly to DataFrame columns.

The lecture demonstrates:

• Mapping numeric category codes to readable names before plotting.
• Using sns.barplot to compare average rentals by working day and season without manually calculating bar offsets.
• Using sns.scatterplot with season as hue to inspect whether the temperature–rental relationship differs by season.
• Using sns.boxplot with weather category and season to examine multiple factors together. Clear-weather days tend to have higher rentals, while rainy or snowy conditions are associated with lower demand.

These plots can reveal group differences and interactions that are hidden by a single overall average.

6. Correlation, heat maps, and regression

A correlation coefficient ranges from −1 to 1:

• Values near 1 indicate a strong positive linear association.
• Values near −1 indicate a strong negative linear association.
• Values near 0 indicate little linear association.

Correlation does not establish causation. Two variables may move together because both respond to another factor.

A correlation matrix can be displayed with sns.heatmap. Options such as annot=True and a color map like coolwarm make the values easier to interpret.

In the bike data, temperature is strongly positively associated with rentals; humidity and wind speed are generally negatively associated. Rentals also correlate strongly with registered-user counts, which is unsurprising because registered users make up a large portion of total rentals.

Correlation is a starting point, not a complete analysis: it can miss nonlinear patterns, conceal differences between groups, and be influenced by outliers. Analysts should inspect plots and investigate unusual observations.

Seaborn’s sns.regplot adds a fitted regression line and a confidence interval to a scatter plot. sns.lmplot can fit separate lines for categories using hue, such as one line per season. These lines summarize trends but assume a linear relationship, so they can be misleading if the true pattern is curved.

7. Practical analysis steps and takeaways

For the monthly-demand exercise, the demonstrated process is:

1. Convert the date column to datetime and then to year-month periods.
2. Group the data by month and calculate the mean rental count.
3. Plot the monthly averages as a labeled line chart, optionally marking each month.
4. Use the smoother series to identify seasonal patterns and support planning.

The exercise finds the highest monthly average in September 2012, at about 7,300 rentals per day. Monthly averages make the seasonal pattern clearer than daily counts and can help managers plan staffing, bike availability, and maintenance.

Overall, choose a visualization based on the question being asked and whether the variables are numerical or categorical. Matplotlib provides detailed control; Seaborn makes many grouped statistical plots quicker to create. The goal is to turn data into evidence that helps answer a real decision-maker’s questions.

Speakers and sources featured

• Speaker: One unnamed instructor or narrator, who explains the material and guides notebook exercises. No other speaker is clearly identified.
• Sources and materials mentioned: Lecture slides; a Jupyter notebook and code demonstrations; the UCI bike-rental dataset; and Python tools including NumPy, Pandas, Matplotlib, Seaborn, requests, and zipfile.