sequifier 0.3.0.8

Train a transformer model with the command line

one-to-one and many-to-one autoregression made easy

Sequifier enables sequence classification or regression for time based sequences using transformer models, via CLI. The specific configuration of preprocessing, which takes a single or multi-variable columnar data file and creates training, validation and test sequences, training, which trains a transformer model, and inference, which calculates model outputs for data (usually the test data from preprocessing), is done via configuration yaml files.

\

Overview

The sequifier package enables:

• the extraction of sequences for training
• the configuration and training of a transformer classification or regression model
• using multiple input and output sequences
• inference on data with a trained model

Other materials

If you want to first get a more specific understanding of the transformer architecture, have a look at the Wikipedia article.

If you want to see a benchmark on a small synthetic dataset with 10k cases, agains a random forest, an xgboost model and a logistic regression, check out this notebook.

Complete example how to build and apply a transformer sequence classifier with sequifier

1. create a conda environment with python >=3.9 activate and run

pip install sequifier

2. create a new project folder (at a path referred to as PROJECT PATH later) and a configs subfolder
3. copy default configs from repository for preprocessing, training and inference
4. adapt preprocess config to take the path to the data you want to preprocess and set project_path toPROJECT PATH
5. run

sequifier preprocess --config_path=[PROJECT PATH]/configs/preprocess.yaml

6. the preprocessing step outputs a "data driven config" at [PROJECT PATH]/configs/ddconfigs/[FILE NAME]. It contains the number of classes found in the data, a map of classes to indices and the oaths to train, validation and test splits of data. Adapt the dd_config parameter in train-on-preprocessed.yaml and infer.yaml in to the path [PROJECT PATH]/configs/ddconfigs/[FILE NAME]and set project_path to PROJECT PATH in both configs
7. run

sequifier train --config_path=[PROJECT PATH]/configs/train-on-preprocessed.yaml

8. adapt data_path in infer.yaml
9. run

sequifier infer --config_path=[PROJECT PATH]/configs/infer.yaml

10. find your predictions at [PROJECT PATH]/outputs/predictions/sequifier-default-best-predictions.csv

More detailed explanations of the three steps

Preprocessing of data into sequences for training

The preprocessing step is designed for scenarios where for timeseries or timeseries-like data, the prediction of the next data point of one or more variables from prior values of these variables and (optionally) other variables is of interest.

This step presupposes input data with three columns: "sequenceId" and "itemPosition", and a column with the variable that is the prediction target. "sequenceId" separates different sequences and the itemPosition column provides values that enable sequential sorting. Often this will simply be a timestamp. You can find an example of the preprocessing input data at documentation/example_inputs/preprocessing_input.csv

The data can then be processed and split into training, validation and testing datasets of all valid subsequences in the original data with the command:

sequifier preprocess --config_path=[CONFIG PATH]

The config path specifies the path to the preprocessing config and the project path the path to the (preferably empty) folder the output files of the different steps are written to.

The default config can be found on this path:

configs/preprocess.yaml

Configuring and training the sequence classification model

The training step is executed with the command:

sequifier train --config_path=[CONFIG PATH]

If the data on which the model is trained DOES NOT come from the preprocessing step, the flag

--on-unprocessed

should be added.

If the training data does not come from the preprocessing step, both train and validation data have to take the form of a csv file with the columns "sequenceId", "subsequenceId", "col_name", [SEQ LENGTH], [SEQ LENGTH - 1],...,"1", "target". You can find an example of the preprocessing input data at documentation/example_inputs/training_input.csv

The training step is configured using the config. The two default configs can be found here:

configs/train.yaml

depending on whether the preprocessing step was executed.

Inferring on test data using the trained model

Inference is done using the command:

sequifier infer --config_path=[CONFIG PATH]

and configured using a config file. The default version can be found here:

configs/infer.yaml