import pandas as pd
import matplotlib.pyplot as plt
from kats.consts import TimeSeriesData
from kats.utils.backtesters import BackTesterSimple
from kats.models.linear_model import LinearModel, LinearModelParams
from kats.models.quadratic_model import QuadraticModel, QuadraticModelParams
from kats.models.arima import ARIMAModel, ARIMAParams
from kats.models.sarima import SARIMAModel, SARIMAParams
from kats.models.theta import ThetaModel, ThetaParams
from kats.models.holtwinters import HoltWintersModel, HoltWintersParams
from kats.models.stlf import STLFModel, STLFParams
from kats.models.prophet import ProphetModel, ProphetParams
# from kats.models.lstm import LSTMModel, LSTMParams # this LSTM implementation is very slow
from kats.models.var import VARModel, VARParams
from kats.models.bayesian_var import BayesianVAR, BayesianVARParams
from kats.utils.backtesters import BackTesterExpandingWindow
import warnings
warnings.simplefilter(action='ignore')
# Record version of key libraries
from importlib.metadata import version
print('kats==%s' % version('kats'))
kats==0.1.0
# Read local data
uni_data_df = pd.read_csv("../data/air_passengers.csv")
multi_data_df = pd.read_csv("../data/multi_ts.csv", index_col=0)
# Set column names so that constructor can recognise the data
# time column is required to be recognised as dates, and any number of value columns can be present
uni_data_df.columns = ['time', 'value']
print(uni_data_df.head())
multi_data_df.columns = ['time', 'v1', 'v2']
print(multi_data_df.head())
time value
0 1949-01-01 112
1 1949-02-01 118
2 1949-03-01 132
3 1949-04-01 129
4 1949-05-01 121
time v1 v2
0 2017-03-12 -0.109 53.8
1 2017-03-13 0.000 53.6
2 2017-03-14 0.178 53.5
3 2017-03-15 0.339 53.5
4 2017-03-16 0.373 53.4
# Convert data object from pandas dataframe into kats object
uni_ts = TimeSeriesData(uni_data_df, time_col_name='time')
multi_ts = TimeSeriesData(multi_data_df, time_col_name='time')
# Visualise data
uni_ts.plot(cols=['value'])
plt.show()
multi_ts.plot(cols=['v1', 'v2'])
plt.show()
Models supported are :
# Test all models available in kats
# Store all models together with parameters. There is no need to fit models first
uni_model_param_dict = {}
# Setup linear model
linear_params = LinearModelParams()
uni_model_param_dict['linear_model'] = (linear_params, LinearModel)
# Setup quadratic model
quadratic_params = QuadraticModelParams()
uni_model_param_dict['quadratic_model'] = (quadratic_params, QuadraticModel)
# Setup arima model
arima_params = ARIMAParams(p=2, d=1, q=1)
uni_model_param_dict['arima_model'] = (arima_params, ARIMAModel)
# Setup sarima model
sarima_params = SARIMAParams(p=2, d=1, q=1)
uni_model_param_dict['sarima_model'] = (sarima_params, SARIMAModel)
# Setup theta model
theta_params = ThetaParams()
uni_model_param_dict['theta_model'] = (theta_params, ThetaModel)
# Setup holtwinters model
hw_params = HoltWintersParams()
uni_model_param_dict['hw_model'] = (hw_params, HoltWintersModel)
# Setup stlf model
stlf_params = STLFParams(method='prophet', m=12)
uni_model_param_dict['stlf_model'] = (stlf_params, STLFModel)
# Setup prophet model
prophet_params = ProphetParams(seasonality_mode='multiplicative')
uni_model_param_dict['prophet_model'] = (prophet_params, ProphetModel)
# Run backtesting for each model
uni_backtesting_errors = {}
for name, obj_tuple in uni_model_param_dict.items():
print('Testing model %s.........' % name)
param, model = obj_tuple
# Setup backtesting object
backtester_obj = BackTesterExpandingWindow(
error_methods=['mape', 'smape', 'mae'],
data=uni_ts,
params=param,
model_class=model,
start_train_percentage=50, #train percentage increases per fold
end_train_percentage=75,
test_percentage=25,
expanding_steps=3 #number of folds
)
# Run backtesting
backtester_obj.run_backtest()
# Store error metric
uni_backtesting_errors[name] = {}
for error, value in backtester_obj.errors.items():
uni_backtesting_errors[name][error] = value
# Convert errors into dataframe
uni_backtesting_err_df = pd.DataFrame.from_dict(uni_backtesting_errors).transpose()
# Sort model ranking by MAPE
uni_backtesting_err_df = uni_backtesting_err_df.sort_values(by='mape', ascending=True)
print(uni_backtesting_err_df)
Testing model linear_model......... Testing model quadratic_model......... Testing model arima_model......... Testing model sarima_model......... Testing model theta_model......... Testing model hw_model......... Testing model stlf_model.........
INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this. INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.
Testing model prophet_model.........
INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this. INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.
mape smape mae prophet_model 0.087252 0.089034 31.986103 stlf_model 0.096418 0.099383 34.872278 arima_model 0.114493 0.121891 46.595001 linear_model 0.117307 0.125707 47.800546 quadratic_model 0.138460 0.138469 53.321068 hw_model 0.172702 0.178684 66.136494 sarima_model 0.174241 0.198778 70.493988 theta_model 0.174968 0.191688 68.784334
# Test all models available in kats
# Store all models together with parameters. There is no need to fit models first
multi_model_param_dict = {}
# Setup var model
var_params = VARParams()
multi_model_param_dict['var_model'] = (var_params, VARModel)
# Setup bayesian var model
bayesvar_params = BayesianVARParams()
multi_model_param_dict['bayesvar_model'] = (bayesvar_params, BayesianVAR)
# # Run backtesting for each model
# multi_backtesting_errors = {}
# for name, obj_tuple in multi_model_param_dict.items():
# print('Testing model %s.........' % name)
# param, model = obj_tuple
# # Setup backtesting object
# # NOTE - this function only supports univariate for now
# backtester_obj = BackTesterExpandingWindow(
# error_methods=['mape', 'smape', 'mae'],
# data=multi_ts,
# params=param,
# model_class=model,
# start_train_percentage=50, #train percentage increases per fold
# end_train_percentage=75,
# test_percentage=25,
# expanding_steps=3 #number of folds
# )
# # Run backtesting
# backtester_obj.run_backtest()
# # Store error metric
# multi_backtesting_errors[name] = {}
# for error, value in backtester_obj.errors.items():
# multi_backtesting_errors[name][error] = value
# # Convert errors into dataframe
# multi_backtesting_err_df = pd.DataFrame.from_dict(multi_backtesting_errors).transpose()
# # Sort model ranking by MAPE
# multi_backtesting_err_df = multi_backtesting_err_df.sort_values(by='mape', ascending=True)
# print(multi_backtesting_err_df)
Testing model var_model.........
--------------------------------------------------------------------------- RemoteTraceback Traceback (most recent call last) RemoteTraceback: """ Traceback (most recent call last): File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/multiprocessing/pool.py", line 125, in worker result = (True, func(*args, **kwds)) File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/kats/utils/backtesters.py", line 310, in _create_model pd.DataFrame( File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/frame.py", line 529, in __init__ mgr = init_dict(data, index, columns, dtype=dtype) File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/internals/construction.py", line 287, in init_dict return arrays_to_mgr(arrays, data_names, index, columns, dtype=dtype) File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/internals/construction.py", line 95, in arrays_to_mgr return create_block_manager_from_arrays(arrays, arr_names, axes) File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/internals/managers.py", line 1706, in create_block_manager_from_arrays raise construction_error(len(arrays), arrays[0].shape, axes, e) File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/internals/managers.py", line 1702, in create_block_manager_from_arrays mgr = BlockManager(blocks, axes) File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/internals/managers.py", line 143, in __init__ self._verify_integrity() File "/home/cylim/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/pandas/core/internals/managers.py", line 323, in _verify_integrity raise construction_error(tot_items, block.shape[1:], self.axes) ValueError: Shape of passed values is (2, 2), indices imply (148, 2) """ The above exception was the direct cause of the following exception: ValueError Traceback (most recent call last) <ipython-input-54-f1d390ca7496> in <module> 20 21 # Run backtesting ---> 22 backtester_obj.run_backtest() 23 24 # Store error metric ~/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/kats/utils/backtesters.py in run_backtest(self) 372 """Executes backtest.""" 373 --> 374 self._build_and_train_models(self._create_train_test_splits()) 375 self.calc_error() 376 ~/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/kats/utils/backtesters.py in _build_and_train_models(self, splits) 366 for train_split, test_split in zip(training_splits, testing_splits) 367 ] --> 368 self.results = [fut.get() for fut in futures] 369 pool.close() 370 ~/anaconda3/envs/experiment_lab/lib/python3.8/site-packages/kats/utils/backtesters.py in <listcomp>(.0) 366 for train_split, test_split in zip(training_splits, testing_splits) 367 ] --> 368 self.results = [fut.get() for fut in futures] 369 pool.close() 370 ~/anaconda3/envs/experiment_lab/lib/python3.8/multiprocessing/pool.py in get(self, timeout) 769 return self._value 770 else: --> 771 raise self._value 772 773 def _set(self, i, obj): ValueError: Shape of passed values is (2, 2), indices imply (148, 2)
# Fit univariate model
prophet_params = ProphetParams(seasonality_mode='multiplicative')
prophet_model = ProphetModel(uni_ts, prophet_params)
prophet_model.fit()
# Fit multivariate model
var_params = VARParams()
var_model = VARModel(multi_ts, var_params)
var_model.fit()
INFO:fbprophet:Disabling weekly seasonality. Run prophet with weekly_seasonality=True to override this. INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.
# Make prediction for next 30 months
uni_fcst = prophet_model.predict(steps=30, freq="MS")
prophet_model.plot()
multi_fcst = var_model.predict(steps=90)
var_model.plot()
plt.show()
