ml-drift-detection 0.1.0

Streamlit dashboard for monitoring data drift and model metrics.

ml-drift-detection 📈🔍

Streamlit dashboard for rapid visual monitoring of data drift and model-performance metrics

---

Why use it?

• One command → full dashboard – point at two files (baseline & new batch) and instantly see
  • Feature-level drift (numeric & categorical)
  • Label drift & concept drift
  • Metric deltas with colour-coded gauges
• Zero vendor lock-in: pure Streamlit + Plotly; runs anywhere Python runs.
• Works as both a command-line tool and an importable library.

---

Default visual settings

Setting	Default value
Background	"white" (pass --background-color="#0E1117" for dark mode)
Gauge colour bands	Relative delta bands for each metric [-1.0 – -0.10] → firebrick [-0.10 – -0.05] → orange [-0.05 – 0.05] → green [ 0.05 – 1.00] → #2ca02c

You can override the bands per metric with --metric-one-threshold-steps,
--metric-two-threshold-steps, etc. Any argument left empty ([]) falls back to these defaults.

---

Installation

pip install ml-drift-detection
# optional dev / test extras
pip install -e ".[dev,test]"

Requires Python ≥ 3.9. See pyproject.toml for full dependency list.

---

Quick-start (CLI)

streamlit run -m ml_drift_detection.cli -- \
  --prod-data=prod_df.csv \
  --new-data=new_df.csv \
  --numeric-cols=age,income,loan_amount \
  --categorical-cols=gender,region,plan \
  --target-variable=subscription_type \
  --target-type=categorical \
  --prod-metrics=accuracy=0.91,f1=0.88 \
  --new-metrics=accuracy=0.83,f1=0.79 \
  --background-color="#0E1117" \
  --metric-one-threshold-steps="[{'range':[-1,-0.1],'color':'red'},{'range':[-0.1,0.05],'color':'orange'},{'range':[0.05,1],'color':'green'}]"

Windows PowerShell multi-line example

streamlit run "C:\Python\Lib\site-packages\ml_drift_detection\cli.py" -- `
  --prod-data="C:\data\prod_df.csv" `
  --new-data="C:\data\new_df.csv" `
  --numeric-cols="age,income,loan_amount" `
  --categorical-cols="gender,region,plan" `
  --target-variable="subscription_type" `
  --target-type="categorical" `
  --prod-metrics="accuracy=0.91,f1=0.88" `
  --new-metrics="accuracy=0.83,f1=0.79"

---

Programmatic use

from ml_drift_detection import dashboard_main
import pandas as pd

prod_df = pd.read_csv("prod_df.csv")
new_df  = pd.read_csv("new_df.csv")

dashboard_main(
    prod_df=prod_df,
    new_df=new_df,
    numeric_cols=["age", "income", "loan_amount"],
    categorical_cols=["gender", "region", "plan"],
    target_variable="subscription_type",
    target_type="categorical",
    prod_metrics={"accuracy": 0.91, "f1": 0.88},
    new_metrics={"accuracy": 0.83, "f1": 0.79},
)

---

Public API (mini)

Symbol	Purpose
ml_drift_detection.cli_main	CLI entry-point (invoked by the command above)
ml_drift_detection.dashboard_main	Programmatic dashboard launcher
check_numeric_stats / check_categorical_stats	Produce pandas summaries of drift metrics
get_plotly_dist / get_plotly_barplot / get_plotly_boxplot	Stand-alone Plotly figure builders

---

Contributing 🤝

git clone https://github.com/knowusuboaky/ml-drift-detection
cd ml-drift-detection
pip install -e ".[dev,test]"
pre-commit install
pytest

Bug reports & feature requests welcome – open an issue first for large changes.

---

License

MIT © Kwadwo Daddy Nyame Owusu-Boakye