: excellent statistical modeling & visualization; tidyverse grammar; strong academic ecosystem. Often preferred for exploratory stats and bespoke plots.
: enterprise analytics; regulatory contexts (e.g., pharma); stable procedures; cloud option via 🔗 flowchart LR
A("explore") -->|run|B("test")
B -->|fix|A
B --> C("explain")
C --> |refactor|A
M02 - Python setup: notebooks, virtual env, conda
CSCI 1109 — Practical Data Science
Other popular data science languages
![]()
: excellent statistical modeling & visualization; tidyverse grammar; strong academic ecosystem. Often preferred for exploratory stats and bespoke plots.
![]()
: enterprise analytics; regulatory contexts (e.g., pharma); stable procedures; cloud option via 🔗 SAS OnDemand for Academics (no local install).- Interoperability: you can mix-and-match (e.g., R ↔︎ Python; SAS OnDemand’s cloud; or export CSV/Parquet).
What is a .ipynb? + modes & shortcuts
- A
.ipynbis a JSON doc: ordered cells (markdown \(\oplus\) code) + outputs + metadata. - Two modes:
- Edit (edit content)
- Command (cell-level actions: render markdown \(\oplus\) execute code)
See 🔗 here
1. Installing Jupyter
From 🔗 here
2. VS Code
- 🔗 VS Code is a popular free, open, multi-platform code editor
- Interpreter selection drives kernel selection in VS Code. Use Python: Select Interpreter and read more 🔗 here
- Jupyter architecture: front end ↔︎ server ↔︎ language kernels. 🔗 more on the back end
3. Google Colab - Notebooks in the cloud
- What it is: a hosted Jupyter environment from Google that runs through your browser.
- Why it’s handy: fast setup, access to free GPUs/TPUs; computation happens in Google’s VM.
- Best for: short-lived demos, workshops, assignments or when your local setup is broken.
- Limits: sessions time out, storage is ephemeral (use Google Drive for persistence), and internet access may be restricted.
- Reproducibility note: good for trying code, not ideal for shipping projects — always export notebooks + environment specs back to local version control.
- 🔗 colab.research.google.com
4. Emacs!
I love 
Emacs, the one true text editor.
As with all things, Jupyter notebooks run in Emacs too, but try this only if you’re playing life on 👷 Hard Mode.
See 🔗 Jupyter Emacs Universe and the 🔗 MELPA package
1. Coffee demand vs temperature
Question: How does outside temperature relate to hot-drink orders (per day)?
Units: temp = °C; orders = count/day.
Code
import pandas as pd, numpy as np
import matplotlib.pyplot as plt
from io import StringIO
coffee = StringIO("""day,temp_C,orders
Mon, 3, 210
Tue, 5, 220
Wed, 7, 265
Thu, 9, 300
Fri,10, 340
Sat, 8, 310
Sun, 4, 250
""")
df = pd.read_csv(coffee)
df['orders_norm'] = df['orders'] / df['orders'].max()
print(df)
plt.figure(figsize=(6,4))
plt.scatter(df['temp_C'], df['orders'], s=80)
plt.title('Coffee orders vs temperature (°C)')
plt.xlabel('Temperature (°C)'); plt.ylabel('Orders / day')
plt.tight_layout(); plt.show() day temp_C orders orders_norm
0 Mon 3 210 0.617647
1 Tue 5 220 0.647059
2 Wed 7 265 0.779412
3 Thu 9 300 0.882353
4 Fri 10 340 1.000000
5 Sat 8 310 0.911765
6 Sun 4 250 0.735294
Interpretation prompt: Is the trend linear? What other variables might confound (weekday, rain, promotions)?
Decision hook: Stock & staffing decisions for chilly vs warm days.
2. Engagement over time
- Let’s measure just visits / day
- Noise (fluctuations) in these measurements can mask the signal (🤩the truth🤩)
- We simulate 2 weeks of daily site visits and use a centered 3-day window.
Units: visits = count/day.
Code
import pandas as pd, numpy as np, matplotlib.pyplot as plt
np.random.seed(0)
days = pd.date_range('2025-01-01', periods=14)
visits = np.random.poisson(lam=np.linspace(80,120,14))
df = pd.DataFrame({'date':days,'visits':visits})
df['roll3'] = df['visits'].rolling(3, center=True).mean()
df['band'] = df['visits'].rolling(3, center=True).std()
print(df.head())
plt.figure(figsize=(7,4))
plt.plot(df['date'], df['visits'], 'o-', label='Daily visits')
plt.fill_between(df['date'],
(df['roll3']-df['band']).fillna(method='bfill').fillna(method='ffill'),
(df['roll3']+df['band']).fillna(method='bfill').fillna(method='ffill'),
alpha=0.3, label='±1 σ (rolling)')
plt.plot(df['date'], df['roll3'], '--', label='3-day mean')
plt.legend(); plt.title('Student engagement (visits/day)')
plt.xlabel('date'); plt.ylabel('visits/day'); plt.tight_layout(); plt.show() date visits roll3 band
0 2025-01-01 81 NaN NaN
1 2025-01-02 86 83.666667 2.516611
2 2025-01-03 84 85.666667 1.527525
3 2025-01-04 87 96.333333 18.823744
4 2025-01-05 118 103.000000 15.524175
Note
Takeaway: Smoothed trend ≠ ground truth; bands show local variability.
Decision hook: When to escalate an intervention vs wait another day.
Mini “open data” example
A tiny excerpt of global CO₂ (MtC = million metric tons of carbon).
Units: MtC/year; this is a toy slice to keep it lightweight.
Code
import pandas as pd, io, matplotlib.pyplot as plt
csv = io.StringIO("""Year,Total_MtC
2015,9724
2016,9737
2017,9908
2018,10004
2019,10010
2020,9469
2021,10081
2022,10302
""")
co2 = pd.read_csv(csv)
print(co2)
co2.plot(x="Year", y="Total_MtC", marker='o', title="Global CO₂ (MtC) — tiny excerpt")
plt.ylabel("MtC / year"); plt.tight_layout(); plt.show() Year Total_MtC
0 2015 9724
1 2016 9737
2 2017 9908
3 2018 10004
4 2019 10010
5 2020 9469
6 2021 10081
7 2022 10302
Discussion: ✏️ What comparisons are valid with such a short series? What would you add (per-capita, sector breakdown, uncertainty)?
