TeremokTSLib 1.2.0

Easy-to-use box ML solution for forcasting consumption

What is this Library about?

Easy-to-use (4 lines of code, actually) framework for training powerful predictive models!

Description

We made a mother-model, which consists of multiple layers of predictive models: ewma is used as trend, Prophet is used for getting seasonality, CatBoost is used for predicting residuals. Why did we do that? Because we needed a out-of-the-box solution, which could be used by non-ML users.

How-to-install?

You can install this framework via pypi:

pip install TeremokTSLib

How-to-use?

You can watch an example in TeremokTSLib/tests foulder. All you need is dataframe with 2 columns: date, consumption. Then you can initiate mother-model and train it with just 2 rows of code:

import TeremokTSLib as tts
model = tts.Model()
model.train(data=data)

Maintained by

Library is developed and being maintained by Teremok ML team

Contacts

• Our website: https://teremok.ru/
• ML team: you can contact us via telegram channel @pivo_txt

Change Log

0.1.0 (27.07.2024)

• First release

1.1.0 (28.07.2024)

• Beta verison release
• Visualisation of itertest added

1.1.1 (06.08.2024)

• Fixed some bugs

1.1.2 (09.08.2024)

• Added parallel training for Prophet

1.1.3 (16.08.2024)

• Now predict_order method returns dict with predicted orders and cons
• Added visualisation of optuna trials

1.1.4 (17.08.2024)

• Optimized Prophet inference. 54% reduction of inference time.

1.1.5 (21.08.2024)

• Fixed bug with Optuna beta optimisation.

1.1.6 (21.08.2024)

• Fixed bug with ewma shift.

1.1.7 - YANKED - (22.08.2024)

• Added regularisation for orders in time of surges in consumption.

1.1.8 (23.08.2024)

• Uploaded fixed seasonality;
• Added WAPE metric calculation in itertest.

1.1.9 (24.08.2024)

• finally fixed beta optimization;
• added regularization parameter to optuna;
• added safe stock coef to optuna;

1.2.0 (24.08.2024)

• added support for lower-than-predicted orders.