The Architecture Diagram

BasicTimeSeries Automation

This part has three classes: TimeSeries, ARIMA and SARIMA classes. TimeSeries is the parent class that takes care of common time series tasks such as:

1. Time Series Plots
2. Decomposition plots
3. Stationarity Checks
4. acf and pacf

• For best results, rigorously clean your data and make sure there are no missing dates
• Note: The data should already have its index set to the time it represents, e.g using

pd.set_index('date_time_column', inplace = True)

And also make sure that the data's datetime variable is in datetime format e.g by doing this:

data["date_column"] = pd.to_datetime(data["date"])
# Or if its yearly data
data["year"] = pd.to_datetime(data["year"], format="%Y")

ARIMA models

ARIMAModels is a class that inherits from the TimeSeries class, with some specific functionalities that include fitting an ARIMA model, checking the performance of the ARIMA model and forecasting using the ARIMA model thats fitted. Below is a limited sample of how to use the ARIMAModels class:

arima_model = ARIMAModels()

Visualizing the trend ,decomposition via decomposition plots and correlation function using acf and pacf

arima_model.trend_visualizations(data)
arima_model.decomposition_plot(data, model='additive', period='None') # The period is None and the model is additive by default. The period should be adjusted according to data. The data can be a pandas DataFrame or a pandas Series.
arima_model.correlation_function(data, lags=10) # lags=10 is the default, and can be changed if need be.

Checking for stationarity

arima_model.stationarity_check(data, signif=0.05) # signif=0.05 is the alpha value, 
# can be changed if need be. In this  case stationarity is being checked at 95% confidence level.

Evaluating the best fitting model

arima_model.model_evaluation(train, test, p_values, d_values, q_values) 
# train parameter should be a Pandas Series, which is the training dataset; 
# test parameter should be a Pandas Series, which is the test dataset; 
# p_values is a list of the range of values that you wish to check for the AR part of ARIMA model
# d_values is a list of the range of values that you wish to check for the difference part
# q_values is a list of the range of values that you wish to check for the MA part of ARIMA model

best_model method fits the best model obtained from the function above and checks timeseries assumptions by comparing the model's prediction and actual test dataset using a line chart. It also checks for serial correlation using Ljung-Box test.

arima_model.best_model(train_data,test_data,order=(2,1,1),lags=10) 
# train_data represents the data to be modelled and should have already been split; 
# order parameter is the order of the ARIMA model.

forecasts method plots a prediction into the future which is crucial and should therefore be as accurate as possible.

arima_model.forecasts(data,order=(2,1,1),steps=20) # data should be the full dataset(neither train nor test)

SARIMA models

SARIMAModels is a class that inherits from the TimeSeries class, with some specific functionalities that include fitting a SARIMA model, checking the performance of the SARIMA model and forecasting using the SARIMA model thats fitted. Below is a limited sample of how to use the SARIMAModels class:

sarima_model = SARIMAModels()

Visualizing the trend ,decomposition via decomposition plots and correlation function using acf and pacf

sarima_model.trend_visualizations(data)
sarima_model.decomposition_plot(data, model='additive', period='None') 
# The period is None and the model is additive by default. 
# The period should be adjusted according to data. 
# The data can be a pandas DataFrame or a pandas Series.
sarima_model.correlation_function(data, lags=10) # lags=10 is the default, and can be changed if need be.

Checking for stationarity

sarima_model.stationarity_check(data, signif=0.05) # signif=0.05 is the alpha value, 
# can be changed if need be. In this  case stationarity is being checked at 95% confidence level.

Evaluating the best fitting model

sarima_model.model_evaluation(train_data, order_limit=2) 
# The data should be the train dataset for best results
# Order limit of 2 ensures that all possible combinations of models upto order 1 will be explored.

best_model method fits the best model obtained from the function above and checks timeseries assumptions by comparing the model's prediction and actual test dataset using a line chart. It also checks for serial correlation using Ljung-Box test.

sarima_model.best_model(train_data, test_data, order=(0, 1, 1),seasonal_order=(1, 1, 1, 12), lags=10) 
# train_data represents the data to be modelled and should have already been split; 
# test_data represents the data to be used for checking the perfomance of the fitted model
# order parameter is the order of the ARIMA part of the model.
# Seasonal_order is the order of the seasonal part of the model.

forecasts method plots a prediction into the future which is crucial and should therefore be as accurate as possible.

sarima_model.forecasts(data,order=(0, 1, 1),seasonal_order=(1, 1, 1, 12),steps=20) 
# data should be the full dataset(neither train nor test)