https://github.com/tblume1992/ConformalImpact

Causal Impact but with MFLES and conformal prediction intervals
https://github.com/tblume1992/ConformalImpact

Last synced: 3 months ago
JSON representation

Causal Impact but with MFLES and conformal prediction intervals

• Host: GitHub
• URL: https://github.com/tblume1992/ConformalImpact
• Owner: tblume1992
• License: mit
• Created: 2024-02-10T17:36:45.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2024-02-10T18:03:29.000Z (9 months ago)
• Last Synced: 2024-07-07T05:01:49.238Z (4 months ago)
• Language: Python
• Size: 273 KB
• Stars: 34
• Watchers: 1
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-conformal-prediction - Conformal Impact

README

        
# Conformal Impact

Take Causal Impact and replace the Bayesian Structural Time Series Model with MFLES and the Basyesian posterior with Conformal Prediction Intervals.

## Quick Examnple an comparison to Causal Impact

```

intervention_effect = 400

np.random.seed(42)

series = np.random.random((130, 1)) * 400

x_series = series * .4 + np.random.random((130, 1)) * 50 + 1000

trend = (np.arange(1, 131)).reshape((-1, 1))

series += 10 * trend

series[-30:] = series[-30:] + intervention_effect

data = pd.DataFrame(np.column_stack([series, x_series]), columns=['y', 'x1'])

import matplotlib.pyplot as plt

plt.plot(series)

plt.plot(x_series)

plt.show()

from ConformalImpact.Model import CI

conformal_impact = CI(opt_size=20,

                      opt_steps=10,

                      opt_step_size=3)

impact_df = conformal_impact.fit(data,

                              n_windows=30,

                              intervention_index=100,

                              seasonal_period=None)

conformal_impact.summary()

conformal_impact.plot()

from causalimpact import CausalImpact

impact = CausalImpact(data, [0, 99], [100, 130])

impact.run()

impact.plot()

print(impact.summary())

output = impact.inferences

np.mean(output['point_effect'].values[-30:])

```