Common Mistakes Beginners Make Learning Data Science

Blank notebook drill (weekly)

1. Pick a dataset you have NOT seen before
2. Close all tutorials
3. In 2 hours, without copying:
   - load and inspect the data
   - answer 3 questions with pandas
   - make 2 charts with clear titles
   - train one simple model with a baseline
4. Afterwards, note every point you had to look up
5. Study those points next week

The list from step 4 is your real syllabus.

import numpy as np

# Why statistics matters: is this difference real or noise?
rng = np.random.default_rng(1)
group_a = rng.normal(100, 15, 30)
group_b = rng.normal(100, 15, 30)   # same true mean!

print(round(group_a.mean() - group_b.mean(), 2))
# A visible difference can appear by chance alone.
# Repeat many times to see how much it varies:
diffs = [rng.normal(100, 15, 30).mean() - rng.normal(100, 15, 30).mean()
         for _ in range(1000)]
print("typical chance difference:", round(np.std(diffs), 2))

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import make_pipeline
from sklearn.linear_model import LogisticRegression

X, y = load_breast_cancer(return_X_y=True)

# WRONG: scaler sees the test data before the split
# X_scaled = StandardScaler().fit_transform(X)
# X_train, X_test, ... = train_test_split(X_scaled, y)

# RIGHT: split first, let a pipeline fit the scaler on training data only
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, stratify=y, random_state=0)
model = make_pipeline(StandardScaler(), LogisticRegression(max_iter=1000))
model.fit(X_train, y_train)
print("test accuracy:", round(model.score(X_test, y_test), 3))

from sklearn.dummy import DummyClassifier
from sklearn.ensemble import RandomForestClassifier

# Continuing with X_train, X_test, y_train, y_test from the left tab
baseline = DummyClassifier(strategy="most_frequent").fit(X_train, y_train)
forest = RandomForestClassifier(random_state=0).fit(X_train, y_train)

print("baseline:", round(baseline.score(X_test, y_test), 3))
print("forest:  ", round(forest.score(X_test, y_test), 3))
# A model is only useful if it clearly beats the baseline.

import numpy as np
from sklearn.metrics import accuracy_score, recall_score, precision_score

# 1000 transactions, only 20 fraud cases
y_true = np.array([1] * 20 + [0] * 980)

# A useless model that always predicts 'not fraud'
y_pred = np.zeros(1000, dtype=int)

print("accuracy: ", accuracy_score(y_true, y_pred))        # 0.98
print("recall:   ", recall_score(y_true, y_pred))           # 0.0
print("precision:", precision_score(y_true, y_pred, zero_division=0))
# 98% accuracy, yet it catches zero fraud.

from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import precision_score, recall_score

X, y = make_classification(n_samples=2000, weights=[0.95], random_state=0)
X_tr, X_te, y_tr, y_te = train_test_split(X, y, stratify=y, random_state=0)
probs = LogisticRegression(max_iter=1000).fit(X_tr, y_tr).predict_proba(X_te)[:, 1]

for t in (0.5, 0.3, 0.1):
    pred = (probs >= t).astype(int)
    print(t, "precision", round(precision_score(y_te, pred, zero_division=0), 2),
             "recall", round(recall_score(y_te, pred), 2))

Data review

[ ] Source, size and time period documented
[ ] Missing values counted and handled
[ ] Duplicates checked
[ ] Target variable distribution examined
[ ] Every feature available at prediction time
[ ] Train/test split done before any fitting
[ ] Stratified split for classification

Model review

[ ] Baseline model reported
[ ] Preprocessing inside a pipeline
[ ] Cross-validation used for model choice
[ ] Test set used only once, at the end
[ ] Metric matches the business problem
[ ] Threshold chosen deliberately
[ ] Errors examined (which cases fail?)

Presentation review

[ ] Problem stated in business terms
[ ] Results compared with baseline
[ ] Limitations listed honestly
[ ] Charts labelled with units
[ ] Next steps proposed
[ ] Code runs from a clean environment
[ ] README explains how to reproduce results