senior-data-scientist

Este extracto es una copia guardada durante la revisión. La fuente externa contiene la versión completa y actual.

---
name: "senior-data-scientist"
description: World-class senior data scientist skill specialising in statistical modeling, experiment design, causal inference, and predictive analytics. Covers A/B testing (sample sizing, two-proportion z-tests, Bonferroni correction), difference-in-differences, feature engineering pipelines (Scikit-learn, XGBoost), cross-validated model evaluation (AUC-ROC, AUC-PR, SHAP), and MLflow experiment tracking — using Python (NumPy, Pandas, Scikit-learn), R, and SQL. Use when designing or analysing controlled experiments, building and evaluating classification or regression models, performing causal analysis on observational data, engineering features for structured tabular datasets, or translating statistical findings into data-driven business decisions.
---

# Senior Data Scientist

World-class senior data scientist skill for production-grade AI/ML/Data systems.

## Core Workflows

### 1. Design an A/B Test

```python
import numpy as np
from scipy import stats

def calculate_sample_size(baseline_rate, mde, alpha=0.05, power=0.8):
    """
    Calculate required sample size per variant.
    baseline_rate: current conversion rate (e.g. 0.10)
    mde: minimum detectable effect (relative, e.g. 0.05 = 5% lift)
    """
    p1 = baseline_rate
    p2 = baseline_rate * (1 + mde)
    effect_size = abs(p2 - p1) / np.sqrt((p1 * (1 - p1) + p2 * (1 - p2)) / 2)
    z_alpha = stats.norm.ppf(1 - alpha / 2)
    z_beta = stats.norm.ppf(power)
    n = ((z_alpha + z_beta) / effect_size) ** 2
    return int(np.ceil(n))

def analyze_experiment(control, treatment, alpha=0.05):
    """
    Run two-proportion z-test and return structured results.
    control/treatment: dicts with 'conversions' and 'visitors'.
    """
    p_c = control["conversions"] / control["visitors"]
    p_t = treatment["conversions"] / treatment["visitors"]
    pooled = (control["conversions"] + treatment["conversions"]) / (control["visitors"] + treatment["visitors"])
    se = np.sqrt(pooled * (1 - pooled) * (1 / control["visitors"] + 1 / treatment["visitors"]))
    z = (p_t - p_c) / se
    p_value = 2 * (1 - stats.norm.cdf(abs(z)))
    ci_low = (p_t - p_c) - stats.norm.ppf(1 - alpha / 2) * se
    ci_high = (p_t - p_c) + stats.norm.ppf(1 - alpha / 2) * se
    return {
        "lift": (p_t - p_c) / p_c,
        "p_value": p_value,
        "significant": p_value < alpha,
        "ci_95": (ci_low, ci_high),
    }

# --- Experiment checklist ---
# 1. Define ONE primary metric and pre-register secondary metrics.
# 2. Calculate sample size BEFORE starting: calculate_sample_size(0.10, 0.05)
# 3. Randomise at the user (not session) level to avoid leakage.
# 4. Run for at least 1 full business cycle (typically 2 weeks).
# 5. Check for sample ratio mismatch: abs(n_control - n_treatment) / expected < 0.01
# 6. Analyze with analyze_experiment() and report lift + CI, not just p-value.
# 7. Apply Bonferroni correction if testing multiple metrics: alpha / n_metrics
```

### 2. Build a Feature Engineering Pipeline

```python
import pandas as pd
import numpy as np
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.impute import SimpleImputer
from sklearn.compose import ColumnTransformer

def build_feature_pipeline(numeric_cols, categorical_cols, date_cols=None):
    """
    Returns a fitted-ready ColumnTransformer for structured tabular data.
    """
    numeric_pipeline = Pipeline([
        ("impute", SimpleImputer(strategy="median")),
        ("scale",  StandardScaler()),
    ])
    categorical_pipeline = Pipeline([
        ("impute", SimpleImputer(strategy="most_frequent")),
        ("encode", OneHotEncoder(handle_unknown="ignore", sparse_output=False)),
    ])
    transformers = [
        ("num", numeric_pipeline, numeric_cols),
        ("cat", categorical_pipeline, categorical_cols),
    ]
    return ColumnTransformer(transformers, remainder="drop")

def add_time_features(df, date_col):
    """Extract