https://github.com/narasimha1997/bzcompute

bzCompute is a computation graph library with built-in support for domain-decomposition and prallel computation. The library can be used for expressing and executing large number of mathematical and text-processing operations using Data-Flow graphs, (Educational version of tensorflow), written in pure python code with numpy support.
https://github.com/narasimha1997/bzcompute

computer-science domain-decomposition educational-project machine-learning parallel-computing

Last synced: 6 days ago
JSON representation

bzCompute is a computation graph library with built-in support for domain-decomposition and prallel computation. The library can be used for expressing and executing large number of mathematical and text-processing operations using Data-Flow graphs, (Educational version of tensorflow), written in pure python code with numpy support.

• Host: GitHub
• URL: https://github.com/narasimha1997/bzcompute
• Owner: Narasimha1997
• License: apache-2.0
• Created: 2019-01-24T05:02:59.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2019-01-27T15:24:47.000Z (almost 6 years ago)
• Last Synced: 2024-10-30T03:37:48.418Z (14 days ago)
• Topics: computer-science, domain-decomposition, educational-project, machine-learning, parallel-computing
• Language: Python
• Size: 258 KB
• Stars: 15
• Watchers: 3
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE.txt

Awesome Lists containing this project

README

        
# bzCompute

bzCompute is a computation graph library with built-in support for domain-decomposition and prallel computation. The library can be used for expressing and executing large number of mathematical and text-processing operations using Data-Flow graphs, (Educational version of tensorflow), written in pure python code with numpy support.

### Examples :

The educational project supports limited number of text-processing and mathematical operations that are normally used frequently.

**Look at examples folder for various examples**

### Parallel Computation : 

The librabry supports both sequential and parallel computation on graphs. Domain Decomposition is the computation technique used to provide task-level parallelism by creating replicas of computation graphs . The library creates a Master-worker setup, where the MasterSession creates specified number of worker sessions, each worker-session is a thread. These threads can be pre-forked to memory or can be created on demand. Each worker is automatically assigned a name. The results are collected. Domain decomposition is manily used for running multiple replicas of data-flow graphs on different inputs, check out *examples*.

### Supported Sessions : 

The library supports three types of sessions as of now : 

  * **SequentialSession** : SequentialSession executes computation graph in a sequential order, leaving parallelism to OS and Hardware.

  *  **MasterSession** : Creates a master session for parallel computation, can be used for domain-decomposition, by specifying number of worker threads.

  *  **StringSession** : It is a variant of SequentialSession for executing a computation graph composed of strign operations. 

 

### Supported Operations : 

bzCompute supports many operations as of now, the library also provides support for defining custom operations as per the requirements.

##### Numerical Operations :

Look at *pyCompute/KernelOperations.py* to obtain list of all supported operations

#### Text-Processing Operations : 

Look at *pyCompute/text_processing/StringOperations.py* for all supported string operations.