kxsystems/automl

Automated Machine Learning Framework for kdb+

Automated machine learning in kdb+

The automated machine learning library described here is built largely on the tools available within the Machine Learning Toolkit. The purpose of this framework is help you automate the process of applying machine learning techniques to real-world problems. In the absence of expert machine-learning engineers this handles the following processes within a traditional workflow.

• Data preprocessing
• Feature engineering and feature selection
• Model selection
• Hyperparameter Tuning
• Report generation and model persistence

Each of these steps is outlined in depth within the documentation. This allows you to understand how decisions are being made and the transformations which their data undergo during the production of the output models.

At present the machine learning frameworks supported for this are based on:

1. One-to-one feature to target non time-series
2. FRESH based feature extraction and model production
3. NLP-based feature creation and word2vec transformation.

The problems which can be solved by this framework will be expanded over time as will the available functionality.

Requirements

The following requirements cover all those needed to run the libraries in the current build of the toolkit.

• embedPy
• ML-Toolkit ≥ 3.0.0

A number of Python dependencies also exist for the running of embedPy functions within both the the machine-learning utilities and FRESH libraries. Install as follows:

pip:

pip install -r requirements.txt

or via conda:

conda install --file requirements.txt

Optional requirements for advanced modules

The above requirements allow you to access the base functionality of AutoML. Additional modules are available – including Sobol sequence hyperparameter search, LaTeX report generation and Keras, PyTorch and NLP models. However, given the large memory requirement for the dependencies of these modules, they are not included in the base functionality and must be installed if required.

Sobol search - via pip (see package details here):

sobol-seq

LaTeX - via conda or pip:

pylatex

Keras - via conda or pip:

keras
tensorflow

PyTorch - via conda or pip:

torch

Theano - via conda or pip:

theano

NLP

The NLP functionality in AutoML requires the KX NLP library along with gensim which can be installed using conda or pip.

Installation

Place the library file in $QHOME and load into a q instance using automl/automl.q

This will load all the functions contained within the .ml namespace

$q automl/automl.q
q).automl.loadfile`:init.q

Documentation

📂 docs

Status

Automated machine learning in kdb+ is still in development and is available here as a beta release, further functionality and improvements will be made to the library in the coming months.

Any issues with the framework should be raised in the issues section of this repository. Functionality suggestions or more general questions should be submitted via email to [email protected]

automl's Issues

'locals error from `.automl.post.report`

When running on kdb+ 3.5 the function .automl.post.report results in a locals error. Function needs to limit the number of local variables to compensate.