7 Myths About Data Science Career Nobody Tells You

import pandas as pd

raw = pd.DataFrame({
    "customer": ["C1", "C2", "C2", "C3", "C4"],
    "city": ["Pune", "pune ", "pune ", "Bengaluru", "Bangalore"],
    "spend": ["1,200", "800", "800", None, "450"],
})

df = raw.drop_duplicates().copy()
df["city"] = (df["city"].str.strip().str.title()
              .replace({"Bangalore": "Bengaluru"}))
df["spend"] = pd.to_numeric(df["spend"].str.replace(",", ""), errors="coerce")
print(df)
print("missing spend:", df["spend"].isna().sum())

import pandas as pd

def profile(df: pd.DataFrame) -> pd.DataFrame:
    """Quick quality summary before any analysis."""
    return pd.DataFrame({
        "dtype": df.dtypes.astype(str),
        "missing_%": (df.isna().mean() * 100).round(1),
        "unique": df.nunique(),
        "example": df.iloc[0],
    })

df = pd.DataFrame({"a": [1, None, 3], "b": ["x", "y", "y"]})
print(profile(df))

# pip install statsmodels
from statsmodels.stats.proportion import proportions_ztest

# Did a new checkout page increase conversion?
conversions = [420, 468]     # old page, new page
visitors = [5000, 5000]

stat, p_value = proportions_ztest(conversions, visitors)
print("old rate:", conversions[0] / visitors[0])
print("new rate:", conversions[1] / visitors[1])
print("p-value:", round(p_value, 4))
# A clear experiment can answer the question without any ML model.

from sklearn.datasets import load_wine
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import cross_val_score
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler

X, y = load_wine(return_X_y=True)
simple = make_pipeline(StandardScaler(), LogisticRegression(max_iter=1000))
complex_ = GradientBoostingClassifier(random_state=0)
for name, m in [("simple", simple), ("complex", complex_)]:
    print(name, cross_val_score(m, X, y, cv=5).mean().round(3))
# If scores are close, the simpler model is often the better choice.

def finding(question, answer, evidence, action, caveat):
    return "\n".join([
        f"Question: {question}",
        f"Answer:   {answer}",
        f"Evidence: {evidence}",
        f"Action:   {action}",
        f"Caveat:   {caveat}",
    ])

print(finding(
    "Why did repeat orders drop last quarter?",
    "Mostly among customers with late first deliveries",
    "Repeat rate 18% (late) vs 34% (on time)",
    "Prioritise delivery speed for first orders",
    "Correlation; confirm with a delivery experiment"))

import numpy as np

def psi(expected, actual, bins=10):
    """Population stability index: a simple drift check."""
    edges = np.quantile(expected, np.linspace(0, 1, bins + 1))
    e = np.histogram(expected, edges)[0] / len(expected)
    a = np.histogram(actual, edges)[0] / len(actual)
    e, a = np.clip(e, 1e-6, None), np.clip(a, 1e-6, None)
    return float(np.sum((a - e) * np.log(a / e)))

rng = np.random.default_rng(0)
train_scores = rng.normal(0.4, 0.1, 5000)
live_scores = rng.normal(0.5, 0.12, 5000)
print("PSI:", round(psi(train_scores, live_scores), 3))

Before touching data, write down:

1. Business question in one sentence
2. Who will use the answer and how
3. Decision that depends on it
4. Data sources and owners
5. What 'good enough' looks like
6. Deadline and check-in dates

[ ] Profiled every table (missing, types, unique)
[ ] Confirmed column meanings with owners
[ ] Checked duplicates and date ranges
[ ] Documented every cleaning decision
[ ] Chose a time window that matches the question
[ ] Noted known biases or gaps

Final write-up structure

- One-line answer
- Key evidence (1-2 charts)
- Recommendation
- Caveats and limitations
- How to repeat or refresh this analysis
- Owner for follow-up and monitoring