AI Roadmap

Stage 1 · Python & Data

40 min readbeginnerView source on GitHub ↗

Data Visualization

Why this matters

df.describe() gives you numbers, but numbers hide shape: two datasets can share the exact same mean and still look completely different once plotted — one clustered tightly, one full of outliers, one with two separate humps. Before you train any model on a dataset, you should be able to look at it. This lesson is a small, permanent toolkit for that — one you’ll reuse for every dataset in the rest of this roadmap.

The concept

When to use each chart type. Most beginner charting mistakes are really “wrong chart for the question,” not a coding error.

Chart Question it answers Example
Line How does one value change in order (usually over time)? Stock price by day
Bar How do totals or averages compare across categories? Average salary by department
Histogram What’s the shape of one numeric column’s values? Distribution of test scores
Scatter How do two numeric columns relate to each other? Years of experience vs. salary

Pick the chart from the question you’re asking, not the other way around.

Matplotlib basics. fig, ax = plt.subplots() gives you a Figure (the whole image) and an Axes (the actual plot area you draw on) — almost every example you’ll see, in this lesson and elsewhere, starts with that line. You draw onto ax (ax.plot(...), ax.bar(...), ax.hist(...), ax.scatter(...)), then either plt.show() to display it interactively or fig.savefig("name.png") to write it to a file. The code examples below use savefig so they run the same way in any environment, including one with no display attached.

Plotting straight from pandas. You rarely need to hand matplotlib raw arrays. A pandas Series or DataFrame has a .plot() method built on top of matplotlib, so avg_salary.plot(kind="bar") draws a bar chart directly from a groupby result in one line, and df.plot.scatter(x=..., y=...) draws a scatter plot straight from two columns. Pass ax=ax to have pandas draw onto an Axes you created yourself, so you can still add labels and a title with matplotlib’s normal methods afterward.

Labelling axes properly. An unlabelled chart is a picture, not information — the reader has to guess what the numbers mean. Every chart you make from now on should call ax.set_xlabel(...), ax.set_ylabel(...), and ax.set_title(...) with real units, not just variable names. “Salary ($)” is a label; “salary” alone, on an axis someone else has to interpret, is not. This is a two-second habit that separates a chart you can hand to someone else from one only you can read.

What a histogram tells you about a feature. A histogram splits a numeric column into equal-width bins and counts how many values fall in each one, so its shape is the shape of your data. A single central hump means most values cluster around one typical value (roughly what your data would look like if it followed a normal, “bell curve” distribution). A long tail stretching to one side means the column is skewed — a handful of very large (or very small) values pulling the mean away from where most of the data actually sits, which is exactly the situation where the median is a more honest summary than the mean. Two separate humps (bimodal) often means you’re actually looking at two different groups mixed into one column — for example, salaries from two very different job levels lumped together. Learning to read these shapes at a glance is most of what “exploring” a dataset means in practice.

Reading a correlation coefficient. A scatter plot shows a relationship visually; a correlation coefficient (what np.corrcoef and pandas’ .corr() compute) puts a number on it, from −1 to +1. A value near +1 means that as one column rises, the other tends to rise too; near −1 means that as one rises, the other tends to fall; near 0 means little to no straight-line relationship between them. As a rough rule of thumb, a magnitude above about 0.7–0.8 is usually considered a strong relationship.

In code

Matplotlib basics — a labelled line plot:

import matplotlib
matplotlib.use("Agg")  # renders to a file with no display window needed
import matplotlib.pyplot as plt

days = [1, 2, 3, 4, 5]
study_minutes = [45, 45, 90, 0, 45]

fig, ax = plt.subplots()
ax.plot(days, study_minutes, marker="o")
ax.set_xlabel("Day")
ax.set_ylabel("Minutes studied")
ax.set_title("Study time, week 1")

fig.savefig("study_time.png")
print("saved study_time.png with", len(days), "points")
print("total minutes:", sum(study_minutes))
saved study_time.png with 5 points
total minutes: 225

A histogram, and what its shape tells you:

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np

rng = np.random.default_rng(seed=42)
test_scores = rng.normal(loc=72, scale=10, size=200)
test_scores = np.clip(test_scores, 0, 100)

fig, ax = plt.subplots()
counts, bin_edges, _ = ax.hist(test_scores, bins=10, edgecolor="white")
ax.set_xlabel("Score")
ax.set_ylabel("Number of students")
ax.set_title("Distribution of test scores")

fig.savefig("test_scores_hist.png")

print("counts per bin:", counts.astype(int))
print("mean score:", round(test_scores.mean(), 1))
print("std dev:", round(test_scores.std(), 1))
counts per bin: [ 6 15 30 40 41 35 19 10  3  1]
mean score: 71.7
std dev: 8.8

The counts climb sharply from 6 up to a peak of 41, then taper off more gradually back down to 1 — a single central hump with a slightly longer tail on the high side. That’s what a roughly-normal distribution centred near 72 looks like as a histogram.

Plotting straight from a pandas groupby result:

import io
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import pandas as pd

csv_text = """name,department,salary,years_experience
Alice,Engineering,95000,5
Bob,Engineering,87000,3
Carol,Sales,72000,4
Dave,Sales,68000,4.5
Eve,Marketing,75000,6
Frank,Marketing,71000,2
Grace,Engineering,102000,8
"""
df = pd.read_csv(io.StringIO(csv_text))

avg_salary = df.groupby("department")["salary"].mean()
print(avg_salary)

fig, ax = plt.subplots()
avg_salary.plot(kind="bar", ax=ax)
ax.set_xlabel("Department")
ax.set_ylabel("Average salary ($)")
ax.set_title("Average salary by department")
fig.tight_layout()
fig.savefig("avg_salary_by_dept.png")
print("saved avg_salary_by_dept.png")
department
Engineering    94666.666667
Marketing      73000.000000
Sales          70000.000000
Name: salary, dtype: float64
saved avg_salary_by_dept.png

A scatter plot, for a relationship between two numeric columns:

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import numpy as np

rng = np.random.default_rng(seed=42)
study_hours = rng.uniform(0, 10, size=50)
test_scores = 50 + 4 * study_hours + rng.normal(0, 8, size=50)

fig, ax = plt.subplots()
ax.scatter(study_hours, test_scores)
ax.set_xlabel("Hours studied")
ax.set_ylabel("Test score")
ax.set_title("Study hours vs test score")
fig.savefig("study_vs_score.png")

correlation = np.corrcoef(study_hours, test_scores)[0, 1]
print("correlation:", round(correlation, 2))
correlation: 0.88

Points trend clearly upward and a correlation of 0.88 confirms it: more study hours are strongly associated with higher scores in this (synthetic) data.

Build this

Using the same employees CSV from the pandas lesson (rebuild it with io.StringIO as in the examples above), make two charts:

  1. A histogram of the salary column. Write one sentence describing its shape — is it clustered, spread out, skewed toward one side?
  2. A scatter plot of years_experience (x-axis) against salary (y-axis). Write one sentence about the relationship you see — does more experience look associated with higher salary here?

Label both charts’ axes and give each a title. Save both to files with fig.savefig(...).

Stretch: compute df["years_experience"].corr(df["salary"]) and compare the number to what your eye told you from the scatter plot — with only 7 rows, does a single correlation number feel trustworthy? Add three more rows of invented data and see whether the correlation and the chart’s shape change much.

Go deeper

Next: Real Datasets