arfx is a family of commandline tools for copying sampled data in and out of ARF containers. ARF (https://github.com/melizalab/arf) is an open, portable file format for storing behavioral and neural data, based on HDF5.

pip install arfx

or from source:

python setup.py install

The general syntax is arfx operation [options] files. The syntax is similar to tar. Operations are as follows:

• -A: copy data from one container to another
• -c: create a new container
• -r: append data to the container
• -t: list contents of the container
• -x: extract entries from the container
• -d: delete entries from the container

Options specify the target ARF file, verbosity, automatic naming schemes, and any metadata to be stored in the entry.

• -f FILE: use ARF file FILE
• -v: verbose output
• -n NAME: name entries sequentially, using NAME as the base
• -a ANIMAL: specify the animal
• -e EXPERIMENTER: specify the experimenter
• -p PROTOCOL: specify the protocol
• -s HZ: specify the sampling rate of the data, in Hz
• -T DATATYPE: specify the type of data
• -u: do not compress data in the arf file
• -P: when deleting entries, do not repack

input files

arfx can read sampled data from pcm, wave, npy and mda files. Support for additional file formats can be added as plugins (see 4).

When adding data to an ARF container (-c and -r modes), the input files are specified on the command line, and added in the order given. By default, entries are given the same name as the input file, minus the extension; however, if the input file has more than one entry, they are given an additional numerical extension. To override this, the -n flag can be used to specify the base name; all entries are given sequential names based on this.

The -n, -a, -e, -p, -s, -T options are used to store information about the data being added to the file. The DATATYPE argument can be the numerical code or enumeration code (run arfx --help-datatypes for a list), and indicates the type of data in the entries. All of the entries created in a single run of arfx are given these values. The -u option tells arfx not to compress the data, which can speed up I/O operations slightly.

Currently only one sampled dataset per entry is supported. Clearly this does not encompass many use cases, but arfx is intended as a simple tool. More specialized import procedures can be easily written in Python using the arf library.

output files

The entries to be extracted (in -x mode) can be specified by name. If no names are specified, all the entries are extracted. All sampled datasets in each entry are extracted as separate channels, because they may have different sampling rates. Event datasets are not extracted.

By default the output files will be in wave format and will have names with the format entry_channel.wav. The -n argument can be used to customize the names and file format of the output files. The argument must be a template in the format defined by the python string module. Supported field names include entry, channel, and index, as well as the names of any HDF5 attributes stored on the entry or channel. The extension of the output template is used to determine the file format. Currently only wave is supported, but additional formats may be supplied as plugins (see 4).

The metadata options are ignored when extracting files; any metadata present in the ARF container that is also supported by the target container is copied.

other operations

As with tar, the -t operation will list the contents of the archive. Each entry/channel is listed on a separate line in path notation.

The -A flag is used to copy the contents of one ARF file to another. The entries are copied without modification from the source ARF file(s) to the target container.

The -d (delete) operation uses the same syntax as the extract operation, but instead of extracting the entries, they are deleted. Because of limitations in the underlying HDF5 library, this does not free up the space, so the file is repacked unless the -P option is set.

The -U (update) operation can be used to add or update attributes of entries, and to rename entries (if the -n flag is set).

The --write-attr operation can be used to store the contents of text files in top-level attributes. The attributes have the name user_<filename>. The --read-attr operation can be used to read out those attributes. This is useful when data collection programs generate log or settings files that you want to store in the ARF file.

This package comes with a few additional scripts that do fairly specific operations.

This script is used to reorganize very large recordings, possibly contained in multiple files, into manageable chunks. Each new entry is given an updated timestamp and attributes from the source entries. Currently, no effort is made to splice data across entries or files. This may result in some short entries. Only sampled datasets are processed.

This script is used to export data into a flat binary structure. It collects sampled data across channels and entries into a single 2-D array. The output can be stored in a multichannel wav file or in a raw binary dat format (N samples by M channels), which is used by a wide variety of spike-sorting tools.

Additional formats for reading and writing can be added using the Python setuptools plugin system. Plugins must be registered in the arfx.io entry point group, with a name corresponding to the extension of the file format handled by the plugin.

An arfx IO plugin is a class with the following required methods:

__init__(path, mode, **attributes): Opens the file at path. The mode argument specifies whether the file is opened for reading (r), writing (w), or appending (a). Must throw an IOError if the file does not exist or cannot be created, and a ValueError if the specified value for mode is not supported. The additional attributes arguments specify metadata to be stored in the file when created. arfx will pass all attributes of the channel and entry (e.g., channels, sampling_rate, units, and datatype) when opening a file for writing. This method may issue a ValueError if the caller fails to set a required attribute, or attempts to set an attribute inconsistent with the data format. Unsupported attributes should be ignored.

read(): Reads the contents of the opened file and returns the data in a format suitable for storage in an ARF file. Specifically, it must be an acceptable type for the arf.entry.add_data() method (see https://github.com/melizalab/arf for documentation).

write(data): Writes data to the file. Must issue an IOError if the file is opened in the wrong mode, and TypeError if the data format is not correct for the file format.

timestamp: A readable property giving the time point of the data. The value may be a scalar indicating the number of seconds since the epoch, or a two-element sequence giving the number of seconds and microseconds since the epoch. If this property is writable it will be set by arfx when writing data.

sampling_rate: A property indicating the sampling rate of the data in the file (or current entry), in units of Hz.

The class may also define the following methods and properties. If any property is not defined, it is assumed to have the default value defined below.

nentries: A readable property indicating the number of entries in the file. Default value is 1.

entry: A readable and writable integer-valued property corresponding to the index of the currently active entry in the file. Active means that the read() and write() methods will affect only that entry. Default is 0, and arfx will not attempt to change the property if nentries is 1.

arfx uses semantic versioning and is synchronized with the major/minor version numbers of the arf package specification.