Capabilities of MIEN

MIEN Primary Features:

• File format support for many file types including:

• Mien internal format (.mien, .mdat)- compressed xml and binary data archive

• USGS SDTS (ident.ddf) - GIS data format

• implemented in an extension block

• uses pullsdts

• Wave (.wav) - common mono and stereo audio format

• xml (.xml, .nmpml) - supports generic XML as well as the specialized nmpml modeling dialect.

• Use of special XML semantic information specified by NeuroML dialects is under development

• Python object persistence formats (pickle and shelve)

• Raw binary data in standard types (eg float64, int16, etc)

• Support for single channel and interleaved multi-channel records

• Neuron hoc (.hoc) - interface language used by the Neuron simulator

• Full support for writing models and experiments to Neuron-readable files

• Read support of hoc written by MIEN, and a subset of general hoc

• No read support for non-declarative hoc code (e.g. functions)

• Swc (.swc) - anatomical data format used by some online anatomy databases

• bbt (.bbt) - binary branched tree anatomical data format

• Neuron Batch Output (.bat) - Read only support to import data written by Neuron simulator batch calls

• Neurolucida ascii (.asc) - Output format from the Neurolucida digitizing software suite

• Matlab v 7 (.mat) - Format used by Matlab

• Nearly complete read/write support

• Suports compressed files

• Doesn't yet support sparse arrays (for now, cast these to "full" for compatible storage)

• DataMAX (.dm) - Output format of RC Electronics data acquisition systems

• Neuron Vector Format (.vec) - An output format used by the Neuron simulator.

• Excel Workbook (.xls) - Unfortunately common spreadsheet format.

• Read support only

• Implemented in an extension block

• uses pyExcellerator

• Ascii and CSV formats (.txt, .csv) - text-based output formats used by spreadsheets, some data loggers, Matlab's dlmread and dlmwrite, etc.

• Image file formats

• supports most common formats, like tiff, png, jpeg, etc

• supports 16bit greyscale tiff

• supports output of sequentially numbered animation frames for easy creation of video with external tools like ImageMagick and Quicktime Pro

• Format conversion between the above formats

• both batch and GUI modes

• Some formats (including mien and mat) can store supersets of data from many file types, eg jpeg, DataMAX, and Neurolucida in the same file.

• Provides reasonable semantic integration for data from different file types, when possible

• General purpose XML infrastructure

• read, write, visualize, and edit any XML

• Incorporate data structures from generic XML into semantically models in nmpml

• The nmpml dialect provides for a unique XPATH for every element, and an ElementReference tag that implements reasonably efficient horizontal links in XML documents.

• Advanced features for particular model classes

• Cell objects provide a wide range of methods useful for editing cell models, measuring complex properties, converting between representations used by different simulation engines, distributing active channels, etc.

• Data objects provide features for handling numerical data, sound, and images

• AbstractModel objects provide features for handling not-declarative content like code snippets and math

• There are many more of these advanced classes

• If there aren't enough, it is relatively easy to add your own

• GUI, CLI, and API interfaces

• Full featured GUIs are provided for exploratory interaction

• Any action that can be done from a GUI can be done from a CLI or within a script

• Scripts can be added to modify GUI menus, allowing rapid macro extensibility

• Adding blocks (GUI extensions scripts) can modify the interface behavior in real time, without reloading the GUI or the data.

• Some of the GUIs can log user actions and auto-generate batch scripts to repeat these actions on more data.

• The DSP GUI is particularly designed for interactive development and testing of a tool chain that will later be used by batch calls, in scripts, or as part of an extension block.

• Advanced visualization classes

• Classes with a high level interface to build 2D, 3D, image, and pseudocolor plots

• Efficient handling of very large plots, including OpenGL hardware acceleration for 3D plots

• Feature-rich interface tools for everything from panning and zooming to interactive selection and editing of 3D objects

• Model building tools

• UIs are provided to integrate support for interactive visualization and selection with support for importing many kinds of data. The result is a powerful set of tools for integrating data from disparate sources into a single model.

• Control of NEURON simulator

• Mien can use Neuron to evaluate compartmental models of cells

• Dispatch of jobs to neuron, and import of the resulting data, is automated

• Mien can provide parallel (and networked) distribution of many independent jobs

• Mien supports modeling network and interaction events with abstract code in Mien, and using these calculations to drive detailed simulations of single cells

• This hybrid model approach is ideal for detailed study of single cell computation in cells that integrate many network inputs.

• Use of external tools in block functions

• Mien can make calls to Matlab or similar computation tools in place of running a python code block in a DSP tool chain or abstract model

• Currently, these calls are inefficient, but they allow users to rapidly develop new algorithms in the language they are most familiar with, and test them in MIEN's interactive environment, and in tool chains that uses MIEN's other built in functions, file support and visualization features, etc.

• Distributed computation

• Mien can distribute tasks efficiently to many CPUs, or many servers in a network environment

• Currently, distribution does not include MPI (or similar technologies), so it is effective when a set of tasks can be computed independently

• Mien can (sometimes) automate separation and distribution of tasks within an optimization scheme or abstract model.

• Optimization and search

• Mien incorporates a general system for interactively building and tuning an abstract model that can subsequently be used as a cost function by a variety of included optimization tools

• Optimization models can include Neuron simulations, abstract python code, Matlab calls, or all of these

• Rapid extensibility

• MIEN blocks provide a framework for extending most aspects of MIEN

• File support, XML classes, GUI behaviors, and abstract model or DSP code can all be added with blocks

• The block interface shields the user from most (though not all) of the internal complexities of the MIEN implementation

• Non-trivial blocks can be very simple. For example, a digital filter takes 2 lines of code to write

• Mien automatically constructs an interface around your block. Windows requesting function parameters, selecting data to operate on, displaying output, etc, are all generated automatically. You never have to write a line of GUI code or touch a GUI pointer or object

• Built-in functions like undo also work on user blocks, without any effort from the block author

• If you don't like MIEN's auto-gui, or need a special interface, you can get access to lower level functions, and write your own interface using MIEN's visualization library, using raw OpenGL code, or even using home-rolled ANSII C. You have the option to get as specific and low level as you like, but 95% of the time you won't need to.