Skip to main content

Advanced data cleaning, data wrangling and feature extraction tools for ML engineers

Project description

Accelerated Machine Learning is a unique approach around machine learning that innovates the data science discovery vertical and productization of the data science delivery model. More specifically, it is an incubator project that shadowed a team of Ph.D. data scientists in connection with the development and delivery of machine learning initiatives to define measurable benefit propositions for customer success. To accomplish this, the project developed specific and unique knowledge regarding transition and preparation of data sets for algorithmic execution and augmented knowledge, which is at the core of the projects services offerings. From this the project developed a new approach to data science discovery and productization dubbed “Accelerated Machine Learning”.

PyPI - Python Version Documentation Status PyPI - License PyPI - Wheel

1   Main features

  • Machine Learning Capability Mapping
  • Parametrized Intent
  • Discovery Transitioning
  • Machine Learning Feature Cataloguing
  • Augmented Knowledge

2   Installation

package install

The best way to install this package is directly from the Python Package Index repository using pip

$ pip install discovery-transition-ds

if you want to upgrade your current version then using pip

$ pip install --upgrade discovery-transition-ds

3   Overview

The Accelerated Machine Learning project is a change of approach in terms of improving productivity of the data scientists. This approach deconstructs the machine learning discovery vertical into a set of capabilities, ideas and knowledge. It presents a completely novel approach to the traditional process automation and model wrapping that is broadly offered as a solution to solve the considerable challenges that currently restrict the effectiveness of machine learning in the enterprise business.

To achieve this, the project offers advanced and specialized programming methods that are unique in approach and novel while maintaining familiarity within common tooling can be identified in four constructs.

1. Machine Learning Capability Mapping - Separation of capabilities, breaking the machine learning vertical into a set of decoupled and targeted layers of discrete and refined actions that collectively present a human lead (ethical AI) base truth to the next set of capabilities. This not only allows improved transparency of, what is, a messy and sometimes confusing set of discovery orientated coded ideas but also loosely couples and targets activities that are, generally, complex and specialized into identifiable and discrete capabilities that can be chained as separately allocated activities.

2. Parametrized Intent - A unique technique extracting the ideas and thinking of the data scientist from their discovery code and capturing it as intent with parameters that can be replayed against productionized code and data. This decoupling and Separation of Concern between data, code and the intent of actions from that code on that data, considerably improves time to market, code reuse, transparency of actions and the communication of ideas between data scientists and product delivery specialists.

3. Discovery Transitioning - Discovery Transitioning - is a foundation of the sepatation of concerns between data provisioning and feature selection. As part of the Accelerated ML discovery Vertical, Transitioning is a foundation base truth facilitating a transparent transition of the raw canonical dataset to a fit-for-purpose canonical dataset to enable the optimisation of discovery analysis and the identification of features-of-interest, for the data scientist and created boundary separation of capabilities decoupling the Data Scientist for the Data Engineer. As output it also provides ‘intelligent Communication’, not only to the Data Scientist through canonical fit-for-purpose datasets, but more generally offers powerful visual discovery tools and artefact generation for production architects, data and business SME’s, Stakeholders and is the initiator of Augmented Knowledge for an enriched and transparent shared view of the extended data knowledge.

4. Machine Learning Feature Cataloguing – With cross over skills within machine learning and advanced data heuristics, investigation identified commonality and separation across customer engagements that particularly challenged our Ph.D data scientists in their effective delivery of customer success. As a result the project designed and developed Feature Cataloguing, a machine learning technique of extracting and engineering features and their characteristics appropriately parameterized for model selection. This technique implements a juxta view of how features are characterized and presented to the modelling layer. Traditionally features are directly mapped as a representation of the underlying data set. Feature Cataloguing treats each individual feature as its own individual set of characteristics as its representation. The resulting outcome considerably improves experimentation, cross feature association, even when unrelated in the original data sets, and the reuse of identified features-of-interest across use case and business domains.

5. Augmented Knowledge - This the ability to capture information on data, activities and the rich stream of subject matter expertise, injected into the machine learning discovery vertical to provide an Augmented n-view of the model build. This includes security, sensitivity, data value scaling, dictionary, observations, performance, optimization, bias, etc. This enriched view of data allows, amongst other things, improved knowledge share, AI explainability, feature transparency, and accountability that feeds into AI ethics, and insight analysis.

4   Background

Born out of the frustration of time constraints and the inability to show business value within a business expectation, this project aims to provide a set of tools to quickly produce visual and observational results. It also aims to improve the communication outputs needed by ML delivery to talk to Pre-Sales, Stakholders, Business SME’s, Data SME’s product coders and tooling engineers while still remaining within familiar code paragigms.

The package looks to build a set of outputs as part of standard data wrangling and ML exploration that, by their nature, are familiar tools to the various reliant people and processes. For example Data dictionaries for SME’s, Visual representations for clients and stakeholders and configuration contracts for architects, tool builders and data ingestion.

4.1   ML Discovery

ML Discovery is first and key part of an end to end process of discovery, productization and tooling. It defines the ‘intelligence’ and business differentiators of everything downstream.

To become effective in the ML discovery phase, the ability to be able to micro-iterate within distinct layers enables the needed adaptive delivery and quicker returns on ML use case.

The building and discovery of an ML model can be broken down into three Separation of Concerns (SoC) or Scope of Responsibility (SoR) for the ML engineer and ML model builder.

  • Data Preparation
  • Feature Engineering
  • Model selection and optimisation

with a forth discipline of insight, interpretation and profiling as an outcome. these three SoC’s can be perceived as eight distinct disciplines

4.2   Conceptuasl Eight stages of Model preparation

  1. Data Loading (fit-for-purpose, quality, quantity, veracity, connectivity)
  2. Data Preparation (predictor selection, typing, cleaning, valuing, validating)
  3. Augmented Knowledge (observation, visualisation, knowledge, value scale)
  4. Feature Attribution (attribute mapping, quantitative attribute characterisation. predictor selection)
  5. Feature Engineering (feature modelling, dirty clustering, time series, qualitative feature characterisation)
  6. Feature Framing (hypothesis function, specialisation, custom model framing, model/feature selection)
  7. Modelling (selection, optimisation, testing, training)
  8. Training (learning, feedback loops, opacity testing, insight, profiling, stabilization)

Though conceptual they do represent a set of needed disciplines and the complexity of the journey to quality output.

4.3   Layered approach to ML

The idea behind the conceptual eight stages of Machine Learning is to layer the preparation and reuse of the activities undertaken by the ML Data Engineer and ML Modeller. To provide a platform for micro iterations rather than a constant repetition of repeatable tasks through the stack. It also facilitates contractual definitions between the different disciplines that allows loose coupling and automated regeneration of the different stages of model build. Finally it reduces the cross discipline commitments by creating a ‘by-design’ set of contracts targeted at, and written in, the language of the consumer.

The concept of being able to quickly run over a single aspect of the ML discovery and then present a stable base for the next layer to iterate against. this micro-iteration approach allows for quick to market adaptive delivery.

5   Getting Started

5.1   First Time Env Setup

When you create a new project, or set up your default master notebook you import the Transition class

from ds_discovery import Transition

Within my master notebook, just as a fail-safe, as it costs nothing, I also set up the environment variable os.environ['DTU_ORIGIN_PATH'] with your root working path. In this example using the environmnt variable of PWD

# set environment variables
os.environ['DTU_ORIGIN_PATH'] = Path(os.environ['PWD']).as_posix()

Along with DTU_ORIGIN_PATH you can also set DTU_PERSIST_PATH and DTU_CONTACT_PATH environment variables allowing one to define where to persist transitioned data and also where to place contract properties. If not set then the origin path is taken as the default for all.

We now have all the appropriate imports and environment variables.

5.2   Transitioning: Data Sourcing

As part of the Accelerated ML Discovery Vertical, Transitioning is a foundation base truth, facilitating a transparent transition of the raw canonical dataset, to a fit-for-purpose canonical dataset, to enable the optimisation of discovery analysis and the identification of features-of-interest. The meaning of cononical is to convert formats into common data language, not just bringing over the dataset but bringing the construct of that dataset ie: type, format, structure, and functionally, in our case because we are Python centric we use Pandas Data Frames as our canonical.

With reference to the diagram, this notebook deals with the Sourcing Contract and the raw canonical dataset as a prerequisite of the Sourcing Contract: 1. Sourcing Notebooks 2. Sourcing Contract 3. Source Connectivity and the Raw Canonical



5.3   Creating a Transitioning Contract Pipeline

  • Creating an instance of the Transitioning Class, passing a unique reference name. when wishing to reference this in other Juptyer Notebooks.
  • The reference name identifies the unique transitioning contract pipeline.
tr = TransitionAgent('synthetic_customer')

5.3.1   Reset the Source Contract

Reset the source contract so we start afresh. Printing the source report validates that our values are empty.

# reset the contract and set the source contract

5.3.2   Find the files

  • Use the discovery find_file(...) to explore the names of the raw files
  • Note, we use the file ‘property manager’ file_pm to get the data_path
  • Because this is a canonical, we can manipulate it as we would our source file
files ='.csv', root_dir=tr.file_pm.data_path).iloc[:,[0,4]].sort_values('name', axis=0)

5.3.3   Build the Source Contract

Source Contract is a set of attributes that define the resource, its type and its location for retrieval and convertion to the raw canonical for transitioning. The Source Contract additionally defines the module and handler that is dynamically loaded at runtime.

By default the source contract requires - resource: a local file, connector, URI or URL - source_type a reference to the type ofresource. if None then extension of resource assumed - location: a path, region or uri reference that can be used to identify location of resource - module_name: a module name with full package path e.g ds_discovery.handlers.pandas_handlers - handler: the name os the handler class - kwargs: additional arguments the handler might require

In this example, because we are using the standard Pandas data frame, file handlers and the localized Transitioning default path locations, as such we only need to provide the resource name and any other Key Word Argument that the specific file handler may need. As our file is csv we have defined the file separator and encoding.

tr.set_source_contract(resource='synthetic_customer.csv', sep=',', encoding='latin1', load=False)   Other Connectivity

As a comparison, in the following example we utilize the vast array of other connectivity options. Here we are looking to connect to an AWS S3 containing csv files .

tr.set_source_contract(resource="synthetic_customer.csv", source_type='s3:csv', location='/sftp/data/repo',
                       module_name='ds_connectivity.handlers.aws_handlers', handler='S3SourceHandler',
                       bucket="synthetic", aws_secret='SaHZedSd4Fd43s/8wDs3/cdsw7cw2',
                       aws_access='DF3E5THA3U85FG', sep=',', encoding='latin1', load=False)

5.4   Source Separation of Concerns

The source details have now been recoreded in the contract pipeline

This Source separation of concerns means: * New Notebooks are no longer tied to the name or location of the data source * File governance and naming convention is managed automatically * Connectivity can be updated or reallocated independantly of the data science activities * Data location and infrastructure, through the delivery lifecycle, can be hardened without effecting the Machine Learning discovery process

5.4.1   Loading the Canonical

Now we have recored the file information, we no longer need to reference these details again To load the contract data we use the transitioning method load_source_canonical() and then we can use the canonical dictionary report to examine the data set.

df = tr.load_source_canonical()

5.4.2   Observations

The report presents our attribute summary as a stylised data frame, highlighting data points of interest. We will see more of this in the next tutorial.

5.4.3   Next Steps

Now we have our raw canonical data extracted and convereted to the canonical from the source we can start the transitioning…

5.5   Python version

Python 2.6 and 2.7 are not supported. Although Python 3.5 is supported, it is recommended to install discovery-transition-ds against the latest Python 3.7.x whenever possible. Python 3 is the default for Homebrew installations starting with version 0.9.4.

5.8   Licence

BSD-3-Clause: LICENSE.

5.9   Authors

Gigas64 (@gigas64) created discovery-transition-ds.

Project details

Release history Release notifications

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for discovery-transition-ds, version 2.3.33
Filename, size File type Python version Upload date Hashes
Filename, size discovery_transition_ds-2.3.33-py37-none-any.whl (59.8 kB) File type Wheel Python version py37 Upload date Hashes View
Filename, size discovery-transition-ds-2.3.33.tar.gz (60.9 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page