Welcome to “Open-Source Tools for Data Science Part 1.” After watching this video, you will be able to: List the open-source data management tools. List the open-source data integration and transformation tools. List the data visualization tools. List the model tools for building, deployment, monitoring, and assessment, and List tools for code and data asset management. So, the most widely used open-source data management tools are relational databases like MySQL and PostgreSQL. Also, there are NoSQL Databases like MongoDB, Apache CouchDB, and Apache Cassandra. In addition, there are file-based tools like the Hadoop File System or Cloud File systems like Ceph. You also have an elastic search tool that stores text data, including the creation of a search index for fast document retrieval. Now, the task of data integration and transformation in the classic data warehousing world is to Extract, Transform, and Load (ETL). Data scientists often propose Extract, Load, Transform (ELT) as data is dumped somewhere, and the data engineer or data scientist handles the transformation of the data. Another term for this process emerged: Data Refinery and Cleansing. The most widely used open-source data integration and transformation tools are the following: Apache AirFlow, which was created by Airbnb originally. KubeFlow, which allows the execution of data science pipelines on top of Kubernetes. Apache Kafka, which originated from LinkedIn. Apache Nifi, which delivers a very nice visual editor. Apache SparkSQL, lets you use ANSI SQL and scales up to compute clusters of thousands of nodes and NodeRED also brings a visual editor. In addition, NodeRED is so low in resource consumption that it even runs on tiny devices like a Raspberry Pi. Now let’s discuss the most widely used open-source data visualization tools. You must distinguish between programming libraries where you must use code or tools containing a user interface. Pixie Dust is also a library but has a user interface that facilitates plotting in Python. A similar approach uses Hue, which can create visualizations from SQL queries. Whereas Kibana, a data exploration, and visualization web application is limited to Elasticsearch (data provider). And finally, Apache Superset is a data exploration and visualization web application. Model deployment is a crucial step. Once you’ve created a machine learning model capable of predicting some critical aspects of the future, you should make it consumable by other developers and turn it into an API. Apache PredictionIO currently only supports Apache Spark ML models for deployment, but support for all libraries is on the roadmap. Seldon is an interesting product since it supports nearly every framework including, TensorFlow, Apache SparkML, R, and scikit learn. Interestingly, it can run on top of Kubernetes and Redhat OpenShift. Another way to deploy SparkML models is MLeap. Finally, TensorFlow can serve any tensor flow model using the TensorFlow service. It can be an embedded device like a Raspberry Pi or smartphone using TensorFlow lite and deployed to a web browser using TensorFlow dot JS. Model monitoring is an important step as well. Once you’ve deployed a machine learning model, you want to track its prediction performance while new data arrives to maintain outdated models. Some examples are the following: ModelDB is a machine model metadata base where information about the models is stored and queried. It natively supports Apache Spark ML Pipelines and scikit-learn. A generic, multi-purpose tool called Prometheus is widely used as well. Although it is not specifically made for machine learning model monitoring, it is used for this purpose. Model performance is measured by more than accuracy. Model bias against protected groups like gender or race is important as well. The IBM AI Fairness 360 open-source toolkit detects and mitigates bias in machine learning models. These models, especially neural network-based deep learning models, can be subject to adversarial attacks where an attacker tries to mislead the model with manipulated data or by controlling it. The IBM Adversarial Robustness 360 Toolbox detects vulnerability against adversarial attacks and leverages the model to be more robust. Finally, machine learning modes are often considered as a black box applying some magic. The IBM AI Explainability 360 toolkit addresses that problem by finding similar examples in a dataset to be presented to an end-user for manual comparison. IBM AI Explainability 360 toolkit can also address the training of a simpler machine learning model to explain the responsibility of different input variables directed toward the final decision of the model. So, the choice of code asset management tools has become quite simple. Git is now the de facto standard for code asset management, also known as version management or version control. Around Git emerged several services. The most prominent is GitHub, but the runner-up is GitLab, with the advantage that the platform is entirely open source and can be hosted and managed on your own. Another choice is Bitbucket. Data asset management, also known as data governance or data lineage, is a crucial part of enterprise-grade data science. Data has to be versioned and annotated with metadata. Apache Atlas is such a tool supporting this task. Another interesting project is ODPi Egeria, managed through the Linux Foundation, is an open ecosystem offers a set of open APIs, types, and interchange protocols that metadata repositories use to share and exchange data. And finally, Kylo is an open-source data management software platform, with extensive support for data asset management tasks. In this video, you learned that: Data management tools are MySQL, PostgreSQL, MongoDB, Apache CouchDB, Apache Cassandra, Hadoop File System, Ceph, and elastic search. Data integration and transformation tools are Apache AirFlow, KubeFlow, Apache Kafka, Apache Nifi, Apache SparkSQL, and NodeRED. Data Visualization tools are Pixie Dust, Hue, Kibana, and Apache Superset. Model deployment tools are Apache PredictionIO, Seldon, Kubernetes, Redhat OpenShift, Mleap, TensorFlow service, TensorFlow lite, and TensorFlow dot JS. Model monitoring tools are ModelDB, Prometheus, IBM AI Fairness 360, IBM Adversarial Robustness 360 Toolbox, and IBM AI Explainability 360. Code asset management tools are Git, GitHub, GitLab, and Bitbucket. And finally, data asset management tools are Apache Atlas, ODPi Egeria, and Kylo.
