Your First Dataset Project, End to End

Every dataset project moves through the same phases: ask a question, find or collect data, explore and clean, analyze, communicate, and share. Miss any phase and the project feels broken. Let's walk through a concrete example end to end.

The example: Spotify top songs

Imagine you want to answer: do songs with higher danceability get more streams? The Kaggle dataset Top Spotify Songs 2023 has exactly what you need. We will take it from download to a polished notebook.

Phase 1: scope your question

• Question: does danceability correlate with streams?
• Success criterion: a single statistic (correlation coefficient) plus a plot
• Time budget: 3 hours total
• Out of scope: causation (which requires experiments)

Phase 2: get the data

# Download from Kaggle (requires kaggle CLI setup) # kaggle datasets download -d nelgiriyewithana/top-spotify-songs-2023 import pandas as pd df = pd.read_csv('spotify-2023.csv', encoding='latin-1') print(df.shape) print(df.columns.tolist()) print(df.head())

Phase 3: explore and clean

# Check types and missing values print(df.dtypes) print(df.isna().sum()) # Streams came in as string with commas, fix it df['streams'] = pd.to_numeric(df['streams'], errors='coerce') df = df.dropna(subset=['streams', 'danceability_%']) # Quick summary print(df[['streams', 'danceability_%']].describe())

Phase 4: analyze

import matplotlib.pyplot as plt from scipy.stats import pearsonr, spearmanr # Pearson assumes linear; Spearman is rank-based pearson_r, p = pearsonr(df['danceability_%'], df['streams']) spearman_r, sp = spearmanr(df['danceability_%'], df['streams']) print(f'Pearson r: {pearson_r:.3f} (p={p:.3f})') print(f'Spearman rho: {spearman_r:.3f} (p={sp:.3f})') # Log-scale the streams for a readable plot plt.scatter(df['danceability_%'], df['streams'], alpha=0.3) plt.yscale('log') plt.xlabel('Danceability (%)') plt.ylabel('Streams (log scale)') plt.title('Danceability vs. Streams') plt.show()

Phase 5: communicate

Phase 6: share

1. 1Push notebook to GitHub or Kaggle
2. 2Write a one-paragraph README with the question, method, finding
3. 3Include a data card noting the dataset source and license
4. 4Tag your notebook with readable headings so others can follow

The big idea: a dataset project is a small disciplined arc. Following the six phases turns chaos into a portfolio piece. Do one, then another, then ten. That is how data intuition develops.