Search

In recent years, both academic research and industry applications see an increased effort in using machine learning methods to measure granular causal effects and design optimal policies based on these causal estimates. Open source packages such as CausalML and EconML provide a unified interface for applied researchers and industry practitioners with a variety of machine learning methods for causal inference. The tutorial will cover the topics including conditional treatment effect estimators by meta-learners and tree-based algorithms, model validations and sensitivity analysis, optimization algorithms including policy leaner and cost optimization. In addition, the tutorial will demonstrate the production of these algorithms in industry use cases.

Modern financial programming and modelling is highly information rich, but our programming tools are often information sparse, especially our strongly typed ones. This leads to an impasse where improvements in programming to aid accuracy and clarity of financial software implementation such as units-of-measure analysis or typed functional programming can't be rolled out to industry. The F# programming language is increasingly used in implementing financial software and models [1]. It includes a feature called “Type Providers” to support the integration of large-scale information sources into a strongly typed functional-first programming environment. In this tutorial, we show that through the use of this simple, intuitive, scalable and on-demand mechanism called "type providers", we can bridge modern strongly-typed languages to the myriad of heterogeneous data services found in the modern financial programming context — for example entity graphs, databases, web services, spreadsheets and even other programming languages. We demonstrate applications using the F# language and the Xenomorph TimeScape time-series database system. • The specific F# 3.0 implementation of the type provider design has the following characteristics: • The mechanism scales to information sources containing extremely large quantities of metadata. • The mechanism can be applied to a myriad of heterogeneous data sources, including databases (SQL, vectored, time-series), web data protocols (WSDL, OData), entity graphs (Freebase), networked data markets and networked information services (World Bank). • The mechanism enables the use of code completion and interactive type checking to increase analyst efficiency when working with rich information sources. • The mechanism interacts with strongly typed tooling such as documentation assistance and completion lists and uses these as the primary way of exploring and understanding the information sources being used. • The programming experience is code focused and integrates well with strongly typed data scripting REPL environments (e.g. F# Interactive). • The programming infrastructure is neutral with regard to the protocol and data formats used. F# itself, as a language, has no specific dependency on external data sources or formats. • The mechanism uses an open architecture, so one can easily add new type providers that consume a different kind of schematized data. • The mechanism integrates technically with advanced features of type systems such as units of measure. Further, it multiplies the value of these features because of the quantity and importance of unitized data available in external in-formation sources. The tutorial will give demonstrations using the F# language. F# is a highly-productive functional-first language for correct, succcinct, robust analytical code in production environments. F# is open source and cross-platform, available under the Apache 2.0 license at http://fsharp.org. The Visual F# Tools are a supported product from Microsoft.

The F# language includes a feature called "F# 3.0 Type Providers" to support the integration of internet-scale information sources into a strongly typed functional-first programming environment. In this position paper we describe the key themes in information-rich functional programming that we have observed during this work. Our contribution is to document these themes and highlight future challenges and opportunities, in the context of a recently released, practical, open-source system for information-rich functional programming. We believe that this area is rich in excellent opportunities for future language and tooling research, information-space integration and schematization techniques.