| commit | 28ff18fca2a9258168db7397d56236a5e0d9564b | [log] [tgz] |
|---|---|---|
| author | Matthias Boehm <mboehm7@gmail.com> | Thu Jul 23 20:11:28 2020 +0200 |
| committer | Matthias Boehm <mboehm7@gmail.com> | Thu Jul 23 20:54:26 2020 +0200 |
| tree | 750fb37f61ee638dda55f98fa373e7d4d1ab9d6d | |
| parent | e93758d639cc22bf8e8193e3164698320664164e [diff] |
[SYSTEMDS-2576] Rework function dictionary (correctness eval/paramserv) This patch fixes a severe correctness issue of second-order functions like eval or paramserv. Specifically, if a script contains a function call (or multiple calls with consistent sizes/scalars) to a function as well as indirect calls with different parameters, our existing IPA too aggressively optimized functions by pushing sizes/scalars into this function. If the same optimized function is then called from eval with completely different parameters, the optimized plan might be invalid, and thus yield incorrect results. In order to get the best of both worlds (optimized functions and the flexibility of second order function calls), we now maintain optimized functions as usual, but if the script contains a second order function, we also keep the unoptimized functions (before inter-procedural analysis) and call these functions as needed. For example, a call from eval, needs to call the unoptimized functions and unoptimized versions of transitively called functions. This further introduces a new function dictionary (that internally maintains both versions of functions), which is used in both function statement blocks and function program blocks. Additionally, this includes several cleanups for the handling of namespaces (removed incomprehensible structure of dml programs) and renamed the function call opcode as we no longer support external functions. As another byproduct, we now have the mechanisms for create deep copies of entire hierarchies of statements (so far we only had that for program blocks).
Overview: SystemDS is a versatile system for the end-to-end data science lifecycle from data integration, cleaning, and feature engineering, over efficient, local and distributed ML model training, to deployment and serving. To this end, we aim to provide a stack of declarative languages with R-like syntax for (1) the different tasks of the data-science lifecycle, and (2) users with different expertise. These high-level scripts are compiled into hybrid execution plans of local, in-memory CPU and GPU operations, as well as distributed operations on Apache Spark. In contrast to existing systems - that either provide homogeneous tensors or 2D Datasets - and in order to serve the entire data science lifecycle, the underlying data model are DataTensors, i.e., tensors (multi-dimensional arrays) whose first dimension may have a heterogeneous and nested schema.
Quick Start Install, Quick Start and Hello World
Documentation: SystemDS Documentation
Python Documentation Python SystemDS Documentation
Status and Build: SystemDS is still in pre-alpha status. The original code base was forked from Apache SystemML 1.2 in September 2018. We will continue to support linear algebra programs over matrices, while replacing the underlying data model and compiler, as well as substantially extending the supported functionalities. Until the first release, you can build your own snapshot via Apache Maven: mvn clean package -P distribution.
