how-to-use.md

Model Configurations

CImpact allows you to configure the underlying models to suit your analysis needs. Below are the configuration options for each supported model.

TensorFlow Model (Bayesian Structural Time Series)

Configure the TensorFlow model using the following parameters:

Parameter	Type	Default	Description
standardize	bool	True	Whether to standardize the data before modeling.
learning_rate	float	0.01	Learning rate for the optimizer.
num_variational_steps	int	1000	Number of steps for variational inference.
fit_method	str	'vi'	Method for fitting the model. Options are 'vi' (Variational Inference) and 'hmc' (Hamiltonian Monte Carlo).

Example Configuration:

model_config = {
    'model_type': 'tensorflow',
    'model_args': {
        'standardize': True,
        'learning_rate': 0.01,
        'num_variational_steps': 1000,
        'fit_method': 'vi'
    }
}

Prophet Model

Configure the Prophet model with the following parameters:

Parameter	Type	Default	Description
standardize	bool	True	Whether to standardize the data before modeling.
Additional Parameters	-	-	Pass any additional parameters supported by Prophet (e.g., seasonality_mode, holidays).

Example Configuration:

model_config = {
    'model_type': 'prophet',
    'model_args': {
        'standardize': True,
        'seasonality_mode': 'multiplicative',
        'weekly_seasonality': True,
        'holidays': your_holidays_dataframe
    }
}

Pyro Model

Configure the Pyro model using the following parameters:

Parameter	Type	Default	Description
standardize	bool	True	Whether to standardize the data before modeling.
learning_rate	float	0.01	Learning rate for the optimizer.
num_iterations	int	1000	Number of iterations for training.
num_samples	int	1000	Number of samples to draw for prediction.

Example Configuration:

model_config = {
    'model_type': 'pyro',
    'model_args': {
        'standardize': True,
        'learning_rate': 0.01,
        'num_iterations': 1000,
        'num_samples': 1000
    }
}

Example Usage

Pyro model

import pandas as pd
from cimpact import CausalImpactAnalysis

# Load your data
data = pd.read_csv('https://raw.githubusercontent.com/Sanofi-Public/CImpact/master/examples/google_data.csv')

# Define the configuration for the model
model_config = {
    'model_type': 'pyro',
    'model_args': {
        'standardize': True,
        'learning_rate': 0.01,
        'num_iterations': 1000,
        'num_samples': 1000
    }
}

# Define the pre and post-intervention periods
pre_period = ['2020-01-01', '2020-03-13']
post_period = ['2020-03-14', '2020-03-31']

#Define index column and target column
index_col = 'date'
target_col = 'y'

# Define color variables
observed_color = "#000000"         # Black for observed
predicted_color = "#7A00E6"        # Sanofi purple for predicted
ci_color = "#D9B3FF66"             # Light lavender with transparency for CI
intervention_color = "#444444"     # Dark gray for intervention
figsize = (10,7)
ci = 95                            # Confidence interval

# Run the analysis
analysis = CausalImpactAnalysis(data, pre_period, post_period, model_config, index_col, target_col, observed_color,  predicted_color, ci_color, intervention_color, ci)
result = analysis.run_analysis()
print(result)

Outcome

Posterior inference {CIMpact}

	Average	Cumulative
Actual	145	2,614
Prediction (s.d.)	130 (718)	2,348 (718)
95% CI	[-2,485, 2,853]	[-2,485, 2,853]
Absolute effect (s.d.)	15 (717)	266 (717)
95% CI	[-1,315, 1,211]	[-1,315, 1,211]
Relative effect (s.d.)	-73.12% (80.40%)	-1316.23% (80.40%)
95% CI	[-161.97%, 192.83%]	[-161.97%, 192.83%]
Posterior tail-area probability p:	0.33167
Posterior probability of a causal effect:	66.83%

*Note: As you can see here, not all models will result into good results! You need to finetune model config to get the best possible result with the model.

Prophet model

import pandas as pd
from cimpact import CausalImpactAnalysis

# Load your data
data = pd.read_csv('https://raw.githubusercontent.com/Sanofi-Public/CImpact/master/examples/google_data.csv')

# Define the configuration for the model
model_config = {
    'model_type': 'prophet',
    'model_args': {
        'standardize': True,
        'learning_rate': 0.01,
        'num_variational_steps': 1000,
        'weekly_seasonality': False,
    }
}

# Define the pre and post-intervention periods
pre_period = ['2020-01-01', '2020-03-13']
post_period = ['2020-03-14', '2020-03-31']

#Define index column and target column
index_col = 'date'
target_col = 'y'

# Define color variables
observed_color = "#000000"         # Black for observed
predicted_color = "#7A00E6"        # Sanofi purple for predicted
ci_color = "#D9B3FF66"             # Light lavender with transparency for CI
intervention_color = "#444444"     # Dark gray for intervention
figsize = (10,7)
ci = 95                            # Confidence interval

# Run the analysis
analysis = CausalImpactAnalysis(data, pre_period, post_period, model_config, index_col, target_col, observed_color,  predicted_color, ci_color, intervention_color, ci)
result = analysis.run_analysis()
print(result)

Outcome

Posterior inference {CIMpact}

	Average	Cumulative
Actual	145	2,614
Prediction (s.d.)	170 (7)	3,064 (7)
95% CI	[142, 195]	[142, 195]
Absolute effect (s.d.)	-25 (14)	-450 (14)
95% CI	[-49, 2]	[-49, 2]
Relative effect (s.d.)	-14.55% (8.08%)	-261.88% (8.08%)
95% CI	[-28.03%, 1.38%]	[-28.03%, 1.38%]
Posterior tail-area probability p:	0.00000
Posterior probability of a causal effect:	100.00%