### Install pandas-datareader

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Installs the pandas-datareader library, redirecting output to /dev/null.

```python
! pip install pandas-datareader > /dev/null
```

--------------------------------

### Import Libraries and Setup Environment

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Imports necessary libraries and configures the environment for Causal Impact analysis. Sets TensorFlow logging level and warning filters.

```python
%matplotlib inline

import sys
import os
import warnings
import logging



logging.getLogger('tensorflow').setLevel(logging.FATAL)
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
warnings.filterwarnings("ignore")
sys.path.append(os.path.abspath('../'))


import numpy as np
import pandas as pd
import tensorflow as tf
import tensorflow_probability as tfp
import matplotlib.pyplot as plt
from causalimpact import CausalImpact

tfd = tfp.distributions
plt.rcParams['figure.figsize'] = [15, 10]
```

--------------------------------

### Install statsmodels

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Installs the statsmodels library, redirecting output to /dev/null to keep the console clean.

```python
! pip install statsmodels > /dev/null
```

--------------------------------

### Install tfcausalimpact

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/README.md

Install the tfcausalimpact package using pip. Ensure you have the required Python version and dependencies like TensorFlow and pandas.

```bash
pip install tfcausalimpact
```

--------------------------------

### Initialize CausalImpact model with pre- and post-intervention periods

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Instantiate the CausalImpact model with the time series data and defined pre-intervention and post-intervention periods. The periods are specified as lists of start and end dates.

```python
pre_period=['2018-01-03', '2020-10-14']
post_period=['2020-10-21', '2020-11-25']

ci = CausalImpact(data, pre_period, post_period)
```

--------------------------------

### Generate Sample Data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

This comment indicates that the following code block is intended to generate sample data, referencing an example from Google's R package for Causal Impact.

```python
# This is an example presented in Google's R code.

```

--------------------------------

### Basic Causal Impact Analysis

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/README.md

Perform a basic causal impact analysis using sample data. This example loads data, defines pre and post intervention periods, initializes CausalImpact, and prints summary reports. It also generates a plot visualizing the results.

```python
import pandas as pd
from causalimpact import CausalImpact


data = pd.read_csv('https://raw.githubusercontent.com/WillianFuks/tfcausalimpact/master/tests/fixtures/arma_data.csv')[['y', 'X']]
data.iloc[70:, 0] += 5

pre_period = [0, 69]
post_period = [70, 99]

ci = CausalImpact(data, pre_period, post_period)
print(ci.summary())
print(ci.summary(output='report'))
ci.plot()
```

--------------------------------

### Model Weekly Seasonality in Daily Data

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

This example demonstrates how to incorporate weekly seasonality into the causal impact model for daily data. It uses `nseasons=7` to model 7 distinct seasonal periods within a week, with each season spanning one day.

```python
import pandas as pd
from causalimpact import CausalImpact

# Daily data with weekly seasonality
data = pd.read_csv('daily_data.csv')
data = data.set_index(pd.date_range(start='20200101', periods=len(data)))

pre_period = ['20200101', '20200311']
post_period = ['20200312', '20200410']

# Model weekly seasonality on daily data (7 seasons, each spanning 1 day)
ci = CausalImpact(
    data,
    pre_period,
    post_period,
    model_args={'nseasons': 7}
)

print(ci.summary())
ci.plot()

# For hourly data with weekly seasonality:
```

--------------------------------

### Custom TensorFlow Probability Models

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Provides an example of using a custom TensorFlow Probability StructuralTimeSeries model with CausalImpact for advanced use cases. Data must be standardized when using custom models with standardize=True.

```python
import numpy as np
import pandas as pd
import tensorflow_probability as tfp
from causalimpact import CausalImpact
from causalimpact.misc import standardize

# Load and prepare data
data = pd.read_csv('data.csv').astype(np.float32)
pre_period = [0, 69]
post_period = [70, 99]

# Standardize data for custom model (required when using custom models with standardize=True)
normed_data, (mu, sig) = standardize(data)
obs_data = normed_data.iloc[:70, 0]  # Pre-intervention response variable
```

--------------------------------

### Causal Impact Analysis with DateTime Index

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

This example shows how to use CausalImpact with a pandas DataFrame that has a datetime index. Periods can be specified using date strings, and the library automatically handles the datetime indexing for analysis.

```python
import pandas as pd
from causalimpact import CausalImpact

# Load data and set datetime index
data = pd.read_csv('comparison_data.csv', index_col=['DATE'])

# Specify periods using date strings
pre_period = ['2019-04-16', '2019-07-14']
post_period = ['2019-07-15', '2019-08-01']

# Run analysis - the library automatically handles datetime indexing
ci = CausalImpact(data, pre_period, post_period)

print(ci.summary())
# Output includes:
# Posterior Inference {Causal Impact}
#                           Average            Cumulative
# Actual                    78574.42           1414339.5
# Prediction (s.d.)         79282.92 (727.48)  1427092.62 (13094.72)
# 95% CI                    [77849.5, 80701.18][1401290.94, 1452621.31]
# ...
```

--------------------------------

### Helper Function to Get Parameter Index

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

A utility function to find the index of a parameter by its name within a model's parameter list. Used when accessing samples from HMC fits.

```python
def get_param_index(model, name):
    for i, v in enumerate(model.parameters):
        if v.name == name:
            return i
```

--------------------------------

### Extracting Sparse Linear Regression Weights (HMC)

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Calculates the mean of the weights for the SparseLinearRegression component when the model is fitted using the HMC method. It uses a helper function to get parameter indices.

```python
tf.reduce_mean(ci.model.components_by_name['SparseLinearRegression/'].params_to_weights(
    ci.model_samples[get_param_index(ci.model, 'SparseLinearRegression/_global_scale_variance')],
    ci.model_samples[get_param_index(ci.model, 'SparseLinearRegression/_global_scale_noncentered')],
    ci.model_samples[get_param_index(ci.model, 'SparseLinearRegression/_local_scale_variances')],
    ci.model_samples[get_param_index(ci.model, 'SparseLinearRegression/_local_scales_noncentered')],
    ci.model_samples[get_param_index(ci.model, 'SparseLinearRegression/_weights_noncentered')],
), axis=0)
```

--------------------------------

### Get Parameter Index for HMC Method

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

A helper function to find the index of a specific parameter within a model's parameter list, used when working with the HMC method in TF Causal Impact.

```python
def get_param_index(model, name):
    for i, v in enumerate(model.parameters):
        if v.name == name:
            return i
    return None
```

--------------------------------

### Initialize Causal Impact Model

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initialize the CausalImpact model with data and pre/post intervention periods. Ensure data is properly formatted.

```python
pre_period = [0, 69]
post_period = [70, 99]
ci = CausalImpact(data, pre_period, post_period)
```

--------------------------------

### Initialize CausalImpact with Hourly Data

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Initializes CausalImpact with hourly data, specifying the pre- and post-intervention periods and model arguments for seasonality.

```python
import pandas as pd

hourly_data = pd.read_csv('hourly_data.csv')
hourly_data = hourly_data.set_index(pd.date_range(start='20200101', periods=len(hourly_data), freq='H'))

ci_hourly = CausalImpact(
    hourly_data,
    ['20200101 00:00:00', '20200311 23:00:00'],
    ['20200312 00:00:00', '20200410 23:00:00'],
    model_args={'nseasons': 7, 'season_duration': 24}
)
```

--------------------------------

### Initialize CausalImpact Model

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initializes the CausalImpact model with specified pre and post-intervention periods. The 'vi' fit method is used for faster fitting.

```python
pre_period = [str(np.min(data.index.values)), "2015-09-13"]
post_period = ["2015-09-20", str(np.max(data.index.values))]

ci = CausalImpact(data.iloc[:, 0], pre_period, post_period, model_args={'nseasons': 52, 'fit_method': 'vi'})
```

--------------------------------

### Generate Summary Output Methods

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Demonstrates how to generate summary outputs from a CausalImpact analysis, including a concise table (default) and a detailed report. Control decimal precision using the 'digits' argument.

```python
import pandas as pd
from causalimpact import CausalImpact

data = pd.read_csv('data.csv')
ci = CausalImpact(data, [0, 69], [70, 99])

# Concise summary table (default)
print(ci.summary())

# Detailed narrative report
print(ci.summary(output='report'))

# Control decimal precision
print(ci.summary(digits=4))
```

--------------------------------

### Fetch and preprocess Bitcoin data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Fetches Bitcoin ('BTC-USD') closing prices from Yahoo Finance for a specified date range. It then removes duplicates, sets the 'Date' as the index, resamples to daily frequency, and fills missing values.

```python
btc_data = pdr.get_data_yahoo(['BTC-USD'], 
                          start=datetime.datetime(2018, 1, 1), 
                          end=datetime.datetime(2020, 12, 3))['Close']
btc_data = btc_data.reset_index().drop_duplicates(subset='Date', keep='last').set_index('Date').sort_index()
btc_data = btc_data.resample('D').fillna('nearest')
```

--------------------------------

### Import necessary libraries

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Imports pandas-datareader for fetching financial data and datetime for date manipulation.

```python
import pandas_datareader as pdr
import datetime
```

--------------------------------

### Initialize CausalImpact with Date Index

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Set the index of your data to a date range and define pre/post intervention periods for CausalImpact analysis. Ensure periods are specified as strings representing dates.

```python
dated_data = data.set_index(pd.date_range(start='20200101', periods=len(data)))

pre_period = ['20200101', '20200311']
post_period = ['20200312', '20200409']

figsize = (20, 6)
ci = CausalImpact(dated_data, pre_period, post_period)
```

--------------------------------

### Initialize CausalImpact with different periods

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Re-initialize the CausalImpact model with a different set of pre- and post-intervention periods to analyze alternative scenarios or timeframes.

```python
pre_period=['2018-01-03', '2018-12-05']
post_period=['2018-12-12', '2019-02-06']

ci = CausalImpact(data, pre_period, post_period)
```

--------------------------------

### Print model sample statistics

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Iterate through the model's posterior samples and print the mean values for different components. This provides insights into the model's learned parameters.

```python
for name, values in ci.model_samples.items():
    print(f'{name}: {values.numpy().mean(axis=0)}')
```

--------------------------------

### Sample and Plot Seasonal State Space Model

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Generate a sample from a defined seasonal state space model and plot the resulting time series data.

```python
plt.plot(s_ssm.sample());
```

--------------------------------

### Python: Perform Causal Impact Analysis

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/README.md

Load data using pandas, define pre- and post-intervention periods, and initialize CausalImpact in Python. Data should be indexed by date.

```python
import pandas as pd
from causalimpact import CausalImpact


data = pd.read_csv('https://raw.githubusercontent.com/WillianFuks/tfcausalimpact/master/tests/fixtures/comparison_data.csv', index_col=['DATE'])
pre_period = ['2019-04-16', '2019-07-14']
post_period = ['2019-7-15', '2019-08-01']
ci = CausalImpact(data, pre_period, post_period, model_args={'fit_method': 'hmc'})
```

--------------------------------

### Initialize Causal Impact with Real Data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initializes the CausalImpact model using real-world data and specified pre- and post-intervention periods. Dates should be in a format recognized by pandas.

```python
pre_period = ['2020-01-01', '2020-03-01']
post_period = ['2020-03-02', '2020-03-31']

ci = CausalImpact(gdata, pre_period, post_period)
```

--------------------------------

### Run Causal Impact Analysis with Default Settings

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

This snippet demonstrates the basic usage of the CausalImpact class, loading data, defining pre and post intervention periods, and printing a summary of the results. Ensure your data is loaded into a pandas DataFrame with response and covariate columns.

```python
import pandas as pd
import numpy as np
from causalimpact import CausalImpact

# Load time series data with response variable (y) in first column and covariates (X) in remaining columns
data = pd.read_csv('https://raw.githubusercontent.com/WillianFuks/tfcausalimpact/master/tests/fixtures/arma_data.csv')[['y', 'X']]

# Simulate an intervention effect starting at index 70
data.iloc[70:, 0] += 5

# Define pre-intervention and post-intervention periods
pre_period = [0, 69]
post_period = [70, 99]

# Run causal impact analysis with default settings
ci = CausalImpact(data, pre_period, post_period)

# Access results
print(ci.summary())
```

--------------------------------

### Initializing Causal Impact with Default Settings

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initializes a CausalImpact object with specified pre and post periods and default model arguments, including standardization.

```python
dated_data = data.set_index(pd.date_range(start='20200101', periods=len(data)))

pre_period = ['20200101', '20200310']
post_period = ['20200311', '20200409']

ci = CausalImpact(dated_data, pre_period, post_period, model_args={'standardize': False})
```

--------------------------------

### Initialize CausalImpact with a Custom Model

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Prepare data by standardizing it and ensuring it has a float32 dtype. Define a custom tfp.sts model, such as a combination of LocalLinearTrend and LinearRegression, and pass it to the CausalImpact constructor. The input data should be the original denormalized data.

```python
from causalimpact.misc import standardize


normed_data, _ = standardize(data.astype(np.float32))

obs_data = normed_data.iloc[:70, 0]

linear_level = tfp.sts.LocalLinearTrend(observed_time_series=obs_data)
linear_reg = tfp.sts.LinearRegression(design_matrix=normed_data.iloc[:, 1:].values.reshape(-1, normed_data.shape[1] -1))

model = tfp.sts.Sum([linear_level, linear_reg], observed_time_series=obs_data)

pre_period = [0, 69]
post_period = [70, 99]

ci = CausalImpact(data, pre_period, post_period, model=model)
```

--------------------------------

### Load and Plot Stock Data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Loads stock data from a CSV file and plots it. Ensure the CSV has 'Date' as the index and is comma-separated.

```python
data = pd.read_csv('../tests/fixtures/volks_data.csv', header=0, sep=' ', index_col='Date', parse_dates=True)
data.plot();
```

--------------------------------

### Configure Causal Impact Model Arguments

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

This snippet illustrates how to customize the structural time series model using the `model_args` parameter. Options include data standardization, prior settings, fitting method ('vi' or 'hmc'), number of posterior samples, and seasonality.

```python
import pandas as pd
from causalimpact import CausalImpact

data = pd.read_csv('data.csv')
pre_period = [0, 69]
post_period = [70, 99]

# Configure model with custom arguments
ci = CausalImpact(
    data,
    pre_period,
    post_period,
    model_args={
        'standardize': True,           # Normalize data to zero mean and unit std (default: True)
        'prior_level_sd': 0.01,        # Prior std for local level; use 0.1 if covariates don't explain y well
        'fit_method': 'vi',            # 'vi' for speed (default), 'hmc' for precision
        'niter': 1000,                 # Number of posterior samples (default: 1000)
        'nseasons': 7,                 # Add weekly seasonality for daily data
        'season_duration': 1           # How many data points each season value spans
    },
    alpha=0.05                         # Significance level for credible intervals (default: 0.05)
)

print(ci.summary())
```

--------------------------------

### Initialize Causal Impact with Covariate

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initializes the CausalImpact model with data that includes a random covariate. 'standardize': False is set to disable data standardization.

```python
pre_period = [0, 69]
post_period = [70, 99]
ci = CausalImpact(data2, pre_period, post_period, model_args={'standardize': False})
```

--------------------------------

### Initializing Causal Impact with HMC Fit Method

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initializes a CausalImpact object with specified pre and post periods and sets the fitting method to Hamiltonian Monte Carlo (HMC). Standardization is also disabled.

```python
ci = CausalImpact(dated_data, pre_period, post_period, model_args={'standardize': False, 'fit_method':'hmc'})
```

--------------------------------

### Data Standardization Utilities

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Utilize `standardize`, `unstandardize`, and `maybe_unstandardize` from `causalimpact.misc` for data preprocessing. `standardize` centers and scales data, while `unstandardize` reverses the process.

```python
import pandas as pd
import numpy as np
from causalimpact.misc import standardize, unstandardize, maybe_unstandardize

# Create sample data
data = pd.DataFrame({
    'y': np.random.randn(100) * 10 + 50,
    'x1': np.random.randn(100) * 5 + 20,
    'x2': np.random.randn(100) * 3 + 10
})

# Standardize data (zero mean, unit standard deviation)
normed_data, (mu, sig) = standardize(data)
print(f"Standardized mean: {normed_data.mean()}")  # ~0
print(f"Standardized std: {normed_data.std()}")    # ~1

# Reverse standardization
original_data = unstandardize(normed_data, (mu, sig))
print(f"Restored mean: {original_data.mean()}")  # Back to original

# Conditional unstandardization (returns data unchanged if mu_sig is None)
result = maybe_unstandardize(normed_data, (mu, sig))  # Returns unstandardized
result = maybe_unstandardize(data, None)               # Returns data unchanged
```

--------------------------------

### Simulate Random Walk with TensorFlow Probability

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Uses TensorFlow Probability to define priors and a LocalLevelStateSpaceModel for simulating a random walk. Requires TensorFlow and TensorFlow Probability.

```python
observed_stddev, observed_initial = (tf.convert_to_tensor(value=1, dtype=tf.float32),
                                     tf.convert_to_tensor(value=0., dtype=tf.float32))
level_scale_prior = tfd.LogNormal(loc=tf.math.log(0.05 * observed_stddev), scale=1, name='level_scale_prior')
initial_state_prior = tfd.MultivariateNormalDiag(loc=observed_initial[..., tf.newaxis], scale_diag=(tf.abs(observed_initial) + observed_stddev)[..., tf.newaxis], name='initial_level_prior')
ll_ssm = tfp.sts.LocalLevelStateSpaceModel(100, initial_state_prior=initial_state_prior, level_scale=level_scale_prior.sample())
```

--------------------------------

### Custom Model with Seasonal Components

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Create a custom CausalImpact model by integrating LocalLinearTrend and Seasonal components for weekly seasonality. Disable internal standardization if data is pre-standardized.

```python
import numpy as np
import pandas as pd
import tensorflow_probability as tfp
from causalimpact import CausalImpact
from causalimpact.misc import standardize

# Prepare data
data = pd.read_csv('data.csv').astype(np.float32)
pre_period = ['2020-01-01', '2020-04-01']
post_period = ['2020-04-02', '2020-05-01']

normed_data, _ = standardize(data)
obs_data = normed_data.loc[:pre_period[1]].iloc[:, 0]

# Build model with multiple components
local_linear = tfp.sts.LocalLinearTrend(observed_time_series=obs_data)
seasonal = tfp.sts.Seasonal(num_seasons=7, observed_time_series=obs_data)  # Weekly seasonality

# Combine components
model = tfp.sts.Sum(
    [local_linear, seasonal],
    observed_time_series=obs_data
)

# Run analysis with custom model (disable internal standardization since we pre-standardized)
ci = CausalImpact(
    data,
    pre_period,
    post_period,
    model=model,
    model_args={'standardize': False}
)

print(ci.summary())
```

--------------------------------

### Initializing Causal Impact Analysis

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initializes the CausalImpact class with the data, pre-intervention period, post-intervention period, and a custom TensorFlow Probability model.

```python
ci = CausalImpact(data, pre_period, post_period, model=model)
```

--------------------------------

### Generate Synthetic Data and Plot

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Generates synthetic data based on the simulated random walk, introduces a structural break, and plots the data using pandas and matplotlib. Ensure pandas and matplotlib are imported.

```python
ll_ssm_sample = np.squeeze(ll_ssm.sample().numpy())

x0 = 100 * np.random.rand(100)
x1 = 90 * np.random.rand(100)
y = 1.2 * x0 + 0.9 * x1 + ll_ssm_sample
y[70:] += 10
data = pd.DataFrame({'x0': x0, 'x1': x1, 'y': y}, columns=['y', 'x0', 'x1'])

data.plot()
plt.axvline(69, linestyle='--', color='k')
plt.legend()
```

--------------------------------

### Run CausalImpact with Seasonal Data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Initialize and run a CausalImpact analysis on data that includes a seasonal component, specifying the number of seasons in the model arguments.

```python
ci = CausalImpact(season_data, pre_period, post_period, model_args={'nseasons': 7})
```

--------------------------------

### Build Custom Model with LocalLinearTrend and LinearRegression

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Construct a custom CausalImpact model by combining LocalLinearTrend and LinearRegression components. Ensure data is pre-standardized if model standardization is disabled.

```python
import numpy as np
import pandas as pd
import tensorflow_probability as tfp
from causalimpact import CausalImpact
from causalimpact.misc import standardize

# Prepare data
data = pd.read_csv('data.csv').astype(np.float32)
pre_period = ['2020-01-01', '2020-04-01']
post_period = ['2020-04-02', '2020-05-01']

normed_data, _ = standardize(data)
obs_data = normed_data.iloc[:, 0]

# Build model with LocalLinearTrend and LinearRegression
linear_level = tfp.sts.LocalLinearTrend(observed_time_series=obs_data)
linear_reg = tfp.sts.LinearRegression(
    design_matrix=normed_data.iloc[:, 1:].values.reshape(-1, normed_data.shape[1] - 1)
)
model = tfp.sts.Sum([linear_level, linear_reg], observed_time_series=obs_data)

# Run causal impact with custom model
ci = CausalImpact(data, pre_period, post_period, model=model)

print(ci.summary())
ci.plot()
```

--------------------------------

### Creating Stationary Data with Log Difference

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Transforms time series data to achieve stationarity by applying a logarithmic transformation followed by differencing. This is a common preprocessing step for time series models.

```python
stationary_data = np.log(data['BTC-USD'].loc[:'2020-10-14']).diff()
```

--------------------------------

### Fetch and preprocess stock market data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Fetches closing prices for multiple stock symbols (TWTR, GOOGL, AAPL, MSFT, AMZN, FB, GOLD) from Yahoo Finance. It processes the data similarly to Bitcoin data: removing duplicates, setting 'Date' as index, resampling to daily, and filling missing values.

```python
X_data = pdr.get_data_yahoo(['TWTR', 'GOOGL', 'AAPL', 'MSFT', 'AMZN', 'FB', 'GOLD'], 
                          start=datetime.datetime(2018, 1, 1), 
                          end=datetime.datetime(2020, 12, 2))['Close']
X_data = X_data.reset_index().drop_duplicates(subset='Date', keep='last').set_index('Date').sort_index()
X_data = X_data.resample('D').fillna('nearest')
```

--------------------------------

### Prepare Data with Random Covariate

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Adds a random covariate 'x2' to the existing data for Causal Impact analysis. This is useful for testing the model's sensitivity to unrelated variables.

```python
x2 = pd.DataFrame(np.random.randn(100, 1), columns=['x2'])
data2 = pd.concat([data, x2], axis=1)
```

--------------------------------

### Plotting time series data with intervention marker

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Visualize the time series data using matplotlib, highlighting a specific intervention point with a vertical dashed line. This helps in visually assessing the data before and after the intervention.

```python
np.log(data).plot(figsize=(15, 12))
plt.axvline('2020-10-14', 0, np.max(data['BTC-USD']), lw=2, ls='--', c='red', label='PayPal Impact')
plt.legend(loc='upper left')
```

--------------------------------

### Concatenate and clean data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Concatenates the Bitcoin data with the other stock market data. It then removes rows with any missing values and resamples the data to a weekly frequency, keeping the last entry of each week and converting to float32.

```python
data = pd.concat([btc_data, X_data], axis=1)
data.dropna(inplace=True)

data = data.resample('W-Wed').last().astype(np.float32)
```

--------------------------------

### Inspect Data Index

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Displays the index of the loaded DataFrame, which should be a DatetimeIndex.

```python
data.index
```

--------------------------------

### Display the full DataFrame

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Print the entire pandas DataFrame to inspect the loaded time series data for all symbols. This is useful for verifying data integrity and structure.

```python
data
```

--------------------------------

### Extracting Sparse Linear Regression Weights (Variational Inference)

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Calculates the mean of the weights for the SparseLinearRegression component when the model is fitted using Variational Inference. This helps assess the contribution of covariates.

```python
tf.reduce_mean(ci.model.components_by_name['SparseLinearRegression/'].params_to_weights(
    ci.model_samples['SparseLinearRegression/_global_scale_variance'],
    ci.model_samples['SparseLinearRegression/_global_scale_noncentered'],
    ci.model_samples['SparseLinearRegression/_local_scale_variances'],
    ci.model_samples['SparseLinearRegression/_local_scales_noncentered'],
    ci.model_samples['SparseLinearRegression/_weights_noncentered'],
), axis=0)
```

--------------------------------

### Analyzing Model Components with TensorFlow Probability

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Decompose the fitted Causal Impact model by component using `tfp.sts.decompose_by_component` and `tfp.sts.decompose_forecast_by_component`. Visualize the mean and standard deviation of each component's contribution.

```python
import numpy as np
import pandas as pd
import tensorflow as tf
import tensorflow_probability as tfp
import matplotlib.pyplot as plt
from causalimpact import CausalImpact

# Run causal impact analysis
data = pd.read_csv('data.csv')
ci = CausalImpact(data, [0, 69], [70, 99], model_args={'standardize': False})

# Decompose fitted time series by component
component_dists = tfp.sts.decompose_by_component(
    ci.model,
    ci.observed_time_series,
    ci.model_samples
)

# Plot each component's contribution
fig, axes = plt.subplots(len(component_dists), 1, figsize=(12, 3 * len(component_dists)))
for ax, (component, component_dist) in zip(axes, component_dists.items()):
    mean = component_dist.mean().numpy()
    stddev = component_dist.stddev().numpy()
    ax.plot(mean, label='Mean')
    ax.fill_between(range(len(mean)), mean - 2*stddev, mean + 2*stddev, alpha=0.3)
    ax.set_title(component.name)
    ax.legend()
plt.tight_layout()
plt.show()

# Decompose forecast by component
forecast_dists = tfp.sts.decompose_forecast_by_component(
    ci.model,
    ci.posterior_dist,
    ci.model_samples
)
```

--------------------------------

### Load Real Data for Causal Impact

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Loads time series data from a CSV file for Causal Impact analysis. Ensure the CSV file has a 'date' column for indexing.

```python
gdata = pd.read_csv('../tests/fixtures/google_data.csv', index_col='date')
```

--------------------------------

### Display data types

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Prints the data types of each column in the processed DataFrame.

```python
data.dtypes
```

--------------------------------

### Add Seasonal Component to Data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Incorporate a sampled seasonal component into the original data to simulate a time series with seasonal patterns for CausalImpact analysis.

```python
season_data = data.copy()
season_data['y'] += np.squeeze(s_ssm.sample().numpy())
```

--------------------------------

### R: Perform Causal Impact Analysis

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/README.md

Load data, define pre- and post-intervention periods, and run the CausalImpact function in R. Ensure data is in zoo format and periods are specified as dates.

```r
data = read.csv.zoo('comparison_data.csv', header=TRUE)
pre.period <- c(as.Date("2019-04-16"), as.Date("2019-07-14"))
post.period <- c(as.Date("2019-07-15"), as.Date("2019-08-01"))
ci = CausalImpact(data, pre.period, post.period)
```

--------------------------------

### Python: Causal Impact with Hamiltonian Monte Carlo

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/README.md

Switch to the Hamiltonian Monte Carlo (HMC) fitting method for higher precision in causal impact analysis. Note that this can significantly increase computation time.

```python
ci = CausalImpact(data, pre_period, post_period, model_args={'fit_method': 'hmc'})
```

--------------------------------

### Display processed stock data

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Displays the DataFrame containing the processed stock market data, showing symbols and their daily closing prices.

```python
X_data
```

--------------------------------

### Plot Forecast Components with CausalImpact

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Use this function to visualize the components of a trained CausalImpact model, such as the forecast and its uncertainty.

```python
plot_forecast_components(ci)
```

--------------------------------

### Yule-Walker Estimation for AR Model Coefficients

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Estimates the coefficients (rho) and noise variance (sigma) of an autoregressive model using the Yule-Walker method. Requires stationary data.

```python
from statsmodels.regression.linear_model import yule_walker

rho, sigma = yule_walker(stationary_data, 1, method='mle')

print(rho, sigma)
```

--------------------------------

### Print Summary of Causal Impact

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Display a concise summary of the causal impact analysis, including actual vs. predicted values, absolute and relative effects, and confidence intervals. This is useful for a quick quantitative overview.

```python
print(ci.summary())
```

--------------------------------

### Use HMC Fitting Method for Higher Precision

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Applies the Hamiltonian Monte Carlo (HMC) fitting method for more accurate posterior sampling, which is slower but yields higher precision. Ensure data is loaded and periods are defined.

```python
import pandas as pd
from causalimpact import CausalImpact

data = pd.read_csv('comparison_data.csv', index_col=['DATE'])
pre_period = ['2019-04-16', '2019-07-14']
post_period = ['2019-07-15', '2019-08-01']

# Use HMC for more precise inference (slower but more accurate)
ci = CausalImpact(
    data,
    pre_period,
    post_period,
    model_args={'fit_method': 'hmc'}
)

# Results will be more precise but computation takes longer
# Consider using GPU acceleration for complex time series
print(ci.summary())
```

--------------------------------

### Plot Time Series Components

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Visualizes the decomposition of the time series into its constituent components based on the fitted Causal Impact model. Requires a fitted CausalImpact object.

```python
def plot_time_series_components(ci):
    # https://github.com/tensorflow/probability/blob/v0.16.0/tensorflow_probability/python/sts/decomposition.py#L165-L195
    component_dists = tfp.sts.decompose_by_component(ci.model, ci.observed_time_series, ci.model_samples)
    num_components = len(component_dists)
    xs = np.arange(len(ci.observed_time_series))
    fig = plt.figure(figsize=(12, 3 * num_components))
    mu, sig = ci.mu_sig if ci.mu_sig is not None else 0.0, 1.0
    for i, (component, component_dist) in enumerate(component_dists.items()):
        # If in graph mode, replace `.numpy()` with `.eval()` or `sess.run()`.
        component_mean = component_dist.mean().numpy()
        component_stddev = component_dist.stddev().numpy()
        ax = fig.add_subplot(num_components, 1, 1 + i)
        ax.plot(xs, component_mean, lw=2)
        ax.fill_between(xs,
                        component_mean - 2 * component_stddev,
                        component_mean + 2 * component_stddev,
                        alpha=0.5)
        ax.set_title(component.name)
        
```

--------------------------------

### Extract Linear Regression Weights (Variational Inference)

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Extracts and prints the mean of covariate weights from a TF Causal Impact model using Variational Inference. Higher absolute values indicate stronger predictive power. Ensure the model has been fitted.

```python
import tensorflow as tf
import pandas as pd
from causalimpact import CausalImpact

data = pd.read_csv('data.csv')
ci = CausalImpact(data, [0, 69], [70, 99], model_args={'standardize': False})

# For Variational Inference (default), samples are in a dictionary
if isinstance(ci.model_samples, dict):
    weights = tf.reduce_mean(
        ci.model.components_by_name['SparseLinearRegression/'].params_to_weights(
            ci.model_samples['SparseLinearRegression/_global_scale_variance'],
            ci.model_samples['SparseLinearRegression/_global_scale_noncentered'],
            ci.model_samples['SparseLinearRegression/_local_scale_variances'],
            ci.model_samples['SparseLinearRegression/_local_scales_noncentered'],
            ci.model_samples['SparseLinearRegression/_weights_noncentered'],
        ),
        axis=0
    )
    print(f"Covariate weights: {weights.numpy()}")
    # Higher absolute values indicate stronger predictive power
```

--------------------------------

### Plot Forecast Components

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Visualizes the decomposition of the forecast into its constituent components. This function is useful for understanding the contribution of each component to the overall forecast. Requires a fitted CausalImpact object.

```python
def plot_forecast_components(ci):
    component_forecasts = tfp.sts.decompose_forecast_by_component(ci.model, ci.posterior_dist, ci.model_samples)
    num_components = len(component_forecasts)
    xs = np.arange(len(ci.post_data))
    fig = plt.figure(figsize=(12, 3 * num_components))
    mu, sig = ci.mu_sig if ci.mu_sig is not None else 0.0, 1.0
    for i, (component, component_dist) in enumerate(component_forecasts.items()):
        # If in graph mode, replace `.numpy()` with `.eval()` or `sess.run()`.
        component_mean = component_dist.mean().numpy()
        component_stddev = component_dist.stddev().numpy()
        ax = fig.add_subplot(num_components, 1, 1 + i)
        ax.plot(xs, component_mean, lw=2)
        ax.fill_between(xs,
                        component_mean - 2 * component_stddev,
                        component_mean + 2 * component_stddev,
                        alpha=0.5)
        ax.set_title(component.name)

```

--------------------------------

### Model Seasonal Components with TensorFlow Probability

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Define a structural time series model with seasonal components using TensorFlow Probability. This involves setting up priors for drift scale and initial state, and specifying the number of seasons and timesteps.

```python
observed_stddev, observed_initial = (5., 0.)
num_seasons = 7

drift_scale_prior = tfd.LogNormal(loc=tf.math.log(.01 * observed_stddev), scale=1.)
initial_effect_prior = tfd.Normal(loc=observed_initial, scale=tf.abs(observed_initial) + observed_stddev)
initial_state_prior = tfd.MultivariateNormalDiag(loc=tf.stack([initial_effect_prior.mean()] * num_seasons, axis=-1),
                                                 scale_diag=tf.stack([initial_effect_prior.stddev()] * num_seasons, axis=-1))

s_ssm = tfp.sts.SeasonalStateSpaceModel(
    num_timesteps=100,
    num_seasons=num_seasons,
    num_steps_per_season=1,
    drift_scale=drift_scale_prior.sample(),
    initial_state_prior=initial_state_prior
)
```

--------------------------------

### Partial Autocorrelation Plot (PACF)

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Plots the partial autocorrelation function (PACF) for a time series. This is crucial for determining the order of an autoregressive (AR) model.

```python
from statsmodels.tsa.stattools import pacf
from statsmodels.graphics.tsaplots import plot_pacf


plot_pacf(stationary_data);
```

--------------------------------

### Plot Seasonal Decomposition Results

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Plots the results of the seasonal decomposition, showing trend, seasonal, and residual components.

```python
result.plot();
```

--------------------------------

### Causal Impact Summary Report

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/README.md

Generate a detailed report for the Causal Impact analysis. This includes posterior inference on actual, predicted, and effect values, along with confidence intervals and probabilities of causal effect.

```python
print(ci.summary(output='report'))
```

--------------------------------

### Seasonal Decomposition of Time Series

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Performs seasonal decomposition on a time series using statsmodels. Ensure the data is preprocessed and the period is correctly specified.

```python
from statsmodels.tsa.seasonal import seasonal_decompose


result = seasonal_decompose(btc_data.loc[:'2020-10-14'].iloc[:, 0])
result.plot();
```

--------------------------------

### Plotting Causal Impact Results

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Visualizes the causal impact analysis using the plot() method, which can display all three panels (original data, pointwise effects, cumulative effects) by default. Customize panels, figure size, and save plots.

```python
import pandas as pd
import matplotlib.pyplot as plt
from causalimpact import CausalImpact

data = pd.read_csv('data.csv')
ci = CausalImpact(data, [0, 69], [70, 99])

# Plot all three panels (default)
ci.plot()

# Plot specific panels only
ci.plot(panels=['original', 'pointwise'])

# Customize figure size
ci.plot(figsize=(15, 12))

# Access figure for further customization (don't show immediately)
ci.plot(show=False)
ax = plt.gca()
ax.set_title('Custom Title')
fig = plt.gcf()
fig.savefig('causal_impact_results.png', dpi=300)
plt.show()
```

--------------------------------

### Plot CausalImpact Results

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Generate a comprehensive plot of the CausalImpact analysis, typically showing the actual data, the predicted values with confidence intervals, and the estimated effects.

```python
ci.plot(figsize=(15, 12))
```

--------------------------------

### Plot Time Series Components

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Plots the components of a Causal Impact analysis. Ensure the CausalImpact object is already defined.

```python
plot_time_series_components(ci)
```

--------------------------------

### Plot Causal Impact Results

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Visualize the results of the CausalImpact analysis, including actual data, predictions, and confidence intervals, using the plot method.

```python
ci.plot(figsize=(15, 14))
```

--------------------------------

### Lag Plot for Time Series

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Generates a lag plot to visualize the correlation between a time series and its lagged values. Useful for identifying patterns and seasonality.

```python
pd.plotting.lag_plot(stationary_data)
```

--------------------------------

### Set Prior for Level Standard Deviation in CausalImpact

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Customize the prior for the standard deviation of the level parameter in CausalImpact. The default is 0.01, suitable for well-behaved data. Increasing this value, like to 0.1, indicates a stronger random walk component not explained by covariates.

```python
ci = CausalImpact(data, pre_period, post_period, model_args={'prior_level_sd':0.1})
```

--------------------------------

### Access Inferred Results

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Access the raw inference results from the fitted CausalImpact model. The `inferences` attribute provides a DataFrame with detailed predictions and confidence intervals, useful for further custom analysis.

```python
ci.inferences.head()
```

--------------------------------

### Select a specific column from the DataFrame

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Access a single column from a pandas DataFrame to isolate a specific time series for analysis. Ensure the column name matches the data.

```python
data['BTC-USD']
```

--------------------------------

### Defining a Causal Impact Model with TensorFlow Probability

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Constructs a structural time series model using TensorFlow Probability, including local linear trend, linear regression, and seasonal components. This model is then used for Causal Impact analysis.

```python
import tensorflow_probability as tfp
from causalimpact.misc import standardize


normed_data, mu_sig = standardize(data)
obs_data = normed_data['BTC-USD'].loc[:'2020-10-14'].astype(np.float32)
design_matrix = pd.concat(
    [normed_data.loc[pre_period[0]: pre_period[1]], normed_data.loc[post_period[0]: post_period[1]]]
).astype(np.float32).iloc[:, 1:]
linear_level = tfp.sts.LocalLinearTrend(observed_time_series=obs_data)
linear_reg = tfp.sts.LinearRegression(design_matrix=design_matrix)
month_season = tfp.sts.Seasonal(num_seasons=4, num_steps_per_season=1, observed_time_series=obs_data, name='Month')
year_season = tfp.sts.Seasonal(num_seasons=52, observed_time_series=obs_data, name='Year')
model = tfp.sts.Sum([linear_level, linear_reg, month_season, year_season], observed_time_series=obs_data)
```

--------------------------------

### Accessing Inference Results

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Retrieves detailed inference results from a CausalImpact object, including predictions, effects, and credible intervals stored in the 'inferences' DataFrame. Also accesses computed attributes like p-value and alpha, and the fitted model.

```python
import pandas as pd
from causalimpact import CausalImpact

data = pd.read_csv('data.csv')
ci = CausalImpact(data, [0, 69], [70, 99])

# Access full inference results as DataFrame
print(ci.inferences.head())

# Access other computed attributes
print(f"P-value: {ci.p_value}")
print(f"Alpha: {ci.alpha}")
print(f"Pre-period data shape: {ci.pre_data.shape}")
print(f"Post-period data shape: {ci.post_data.shape}")

# Access the fitted model and samples for advanced analysis
model = ci.model
samples = ci.model_samples
```

--------------------------------

### Working with NumPy Arrays

Source: https://context7.com/willianfuks/tfcausalimpact/llms.txt

Perform Causal Impact analysis using NumPy arrays as input. The library automatically converts arrays to DataFrames internally. Ensure the first column is the response variable.

```python
import numpy as np
from causalimpact import CausalImpact

# Generate synthetic data as numpy array
np.random.seed(42)
n_points = 100

# Response variable with intervention effect
y = np.cumsum(np.random.randn(n_points)) + 100
y[70:] += 10  # Add intervention effect

# Covariates
x1 = np.cumsum(np.random.randn(n_points)) + 50
x2 = np.cumsum(np.random.randn(n_points)) + 30

# Combine into array (first column is response, rest are covariates)
data = np.column_stack([y, x1, x2])

# Run causal impact with numpy array
pre_period = [0, 69]
post_period = [70, 99]

ci = CausalImpact(data, pre_period, post_period)

print(ci.summary())
ci.plot()
```

--------------------------------

### Plot Specific Causal Impact Panels

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Customize the plot to display only specific panels, such as 'original' and 'pointwise' effects. Adjust figsize for better visualization.

```python
ci.plot(panels=['original', 'pointwise'], figsize=(15, 12))
```

--------------------------------

### Plot Causal Impact with Specific Panels

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Generates a plot for the Causal Impact analysis, displaying only the 'original' data panel. Useful for focusing on the raw time series.

```python
ci.plot(figsize=(15, 6), panels=['original'])
```

--------------------------------

### Causal Impact with Linear Regression and Data Holes

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Performs causal impact analysis on time series data with missing values (holes) by applying regularization and filling NaNs. This approach ensures proper inference when using linear regression models.

```python
from causalimpact.misc import standardize


pre_period = ['20200101', '20200311']
post_period = ['20200312', '20200409']

dated_data_holes = dated_data[dated_data.index.isin(['20200315', '20200316', '20200404']) == False]
reg_dated_data_holes = tfp.sts.regularize_series(dated_data_holes)
normed_data = standardize(reg_dated_data_holes.astype(np.float32))[0]
obs_data = normed_data.loc[pre_period[0]: pre_period[1]].iloc[:, 0]

reg_normed_data = tfp.sts.regularize_series(normed_data)
design_matrix_data = (reg_normed_data.iloc[:, 1:]).astype(np.float32).fillna(0)

linear_level = tfp.sts.LocalLinearTrend(observed_time_series=obs_data)
linear_reg = tfp.sts.LinearRegression(design_matrix=design_matrix_data.values.reshape(-1, normed_data.shape[1] -1))

model = tfp.sts.Sum([linear_level, linear_reg], observed_time_series=obs_data)

ci = CausalImpact(dated_data_holes, pre_period, post_period, model=model)
```

--------------------------------

### Autocorrelation Plot for Time Series

Source: https://github.com/willianfuks/tfcausalimpact/blob/master/notebooks/getting_started.ipynb

Generates an autocorrelation plot (ACF) to visualize the correlation between a time series and its lagged values. Helps in identifying the order of AR or MA models.

```python
pd.plotting.autocorrelation_plot(stationary_data);
```