Jupyter Python Data Science Expert-AI Python data science assistant

Example

```python # quick EDA pattern inside a notebook import pandas as pd import seaborn as sns import matplotlib.pyplot as plt df = pd.read_csv('transactions.csv') print(df.info()) print(df.describe()) # missingness overview sns.heatmap(df.isna(), cbar=False) # pairwise relationships on a sample sns.pairplot(df.sample(500), hue='customer_segment') plt.show() ```

Scenario

A retail analyst receives a 10M-row transactions CSV. Inside a Jupyter notebook they sample rows, inspect data distributions, identify high-cardinality categorical columns, discover that 12% of 'purchase_amount' is missing, and iterate on imputation strategies. Visualizations (histograms, boxplots, time-series aggregates) help decide feature engineering—e.g., log-transforming skewed amounts and creating rolling-average features for recency-frequency analysis.

Example

```python # prototyping a pipeline + CV AUC from sklearn.pipeline import make_pipeline from sklearn.impute import SimpleImputer from sklearn.preprocessing import StandardScaler from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import cross_val_score pipe = make_pipeline( SimpleImputer(strategy='median'), StandardScaler(), RandomForestClassifier(n_estimators=200, random_state=42) ) scores = cross_val_score(pipe, X, y, cv=5, scoring='roc_auc') print('mean CV AUC:', scores.mean()) ```

Scenario

A data scientist building a churn model iterates through feature sets and model families (XGBoost, RandomForest, logistic regression) within a notebook. They run repeated cross-validation, log metrics and feature importances to the notebook, then use SHAP inside the same environment to produce explanation plots. They refine preprocessing inside a scikit-learn Pipeline so the same transformations can be exported and used in production code.

Example

```bash # parameterize and run a notebook for a specific window papermill analysis.ipynb analysis_out.ipynb -p START_DATE '2025-01-01' -p END_DATE '2025-06-30' # export to HTML for stakeholders jupyter nbconvert --to html analysis_out.ipynb # serve an interactive dashboard built from a notebook voila dashboard.ipynb --port=8866 ```

Scenario

A data science team delivers a weekly report: they parameterize a master notebook with papermill to accept the week-start and week-end, run it in CI to produce an HTML report and a JSON summary for ingestion into a BI system. For an internal interactive dashboard, they expose widgets and serve it via Voila. For productionizing a model, they extract the preprocessing pipeline into a Python package, containerize the runtime with Docker (pinned dependencies via environment.yml), and add a CI job that runs the notebook end-to-end using nbclient as an integration test.