v0.4.4 - Timing and Sound (February 2nd, 2022)¶
Several parts to this update!
See PR#146 for details about improvements to jackd sound timing! In short:
Changed the way that continuous sounds work. Rather than cycling through an array, which was easy to drop, now pass a sound object that can generate its own samples on the fly using the hydration module.
More accurate timing of sound ending callbacks. Before, the event would be called immediately on buffering the sounds into the jack ports, but that was systematically too early. Instead, use jack timing methods to account for delay from blocksize and n_periods to wait_until a certain delay to set() the event. See _wait_for_end
hydrationmodule for creating and storing autopilot objects between processes and computers!
@Introspectmade and added to sound classes. Will be moved to root class. Allows storing the parameters given on instantiation.
requiresmodule for more explicit declarations of by-object dependencies to resolve lots of the little fragile checks throughout the package, as well as make it easier for plugins :)
typesmodule that will, well, have types for v0.5.0’s reworked type system!
minor - added exceptions module, just stubs for now
Made dummy sound class to just use sounds without needing a running sound server
New transformations! The Prasad line fitting algorithm as
Linefit_Prasadand ordering points in a line from, eg. edge detection in ``Order_Points` `
Only warn once for returning a default pref value, and make its own warning class so that it can be filtered.
Cleaning up the base sound classes and moved them to their own module because sounds was very cumbersome and hard to reason about. Now use
get_sound_classinstead of declaring within the module.
Made optional install packages as
extras_requireso now can install with
pip install auto-pi-lot -E pilotrather than autodetecting based on architecture. Further improvements (moving to poetry) will be in v0.5.0
Correctly identify filenames in logging, before the last module name was treated as a suffix on the path and removed, and so only the most recent logger created would actually log to disk. Logging now works across threads and processes.
Fall back to a non-multiprocessing-based prefs if for some reason we can’t use a mp.Manager in the given context (eg. ipython) - Still need to figure out a way to not print the exception because it is thrown asynchronously.
as much as i love it, the splash screen being absent for whatever reason shouldn’t crash the program.
Raise an exception when instantiating a picamera without having picamera installed, re: https://github.com/wehr-lab/autopilot/issues/142
Raise ImportError when ffmpeg is not present and trying to use a videowriter class
Use a deque rather than an infinitely growing list to store GPIO events.
Documenting the scripts module a bit better.
Lots more docs on jack_server
v0.4.3 (October 20th, 2021)¶
579ef1a - En route to implementing universal calibrations, load and save them in a specified place for each hardware object instead of the horrific olde way which was built into
prefsfor some reason
prefs attempts to make directories if they don’t exist
plenty of new debugging flags!
a775723 - Sleep before graduating tasks, lateral fix until we rework the task initiation ritual
360062d - pad sounds with silence or continuous sounds if they aren’t a full period length
6614c80 - Revert to old way of making chunks to make it work with both padded and unpadded sounds
Import sounds module directly instead of referring from the package root in tests
Terminal node pings pilots instead of an erroneous reference to a nonexistent
47dd4c2 - Fix pinging by passing pilot id, and handle pressing start/stop button when subject not selected
Fixed some GUI exceptions from trying to make blank lines in reassign window, improperly handling the Subject class.
v0.4.2 (August 24th)¶
Fixed a circular import problem that prevented the stim module from being imported because the placeholder metaclass was in the __init__.py file. Moved it to its own file.
Fixed another instantiated but not raised value error in gpio
Documenting flags in networking objects
Documenting min_size in camera stream method
Documenting invert_gyro in I2C_9DOF
v0.4.1 (August 17th)¶
autopilot.setup.forms.HARDWARE_FORMwould incorrectly use the class object itself rather than the class name in a few places which caused hardware names to incorrectly display and be impossible to add!
Correctly handle module name in loggers when running interactively
Use accelerometer calibration when computing
Document the attributes in
v0.4.0 - Become Multifarious (August 3rd, 2021)¶
This release is primarily to introduce the new plugin system, the autopilot wiki, and their integration as a way of starting the transformation of Autopilot into a tool with decentralized development and governance (as well as make using the tool a whole lot easier and more powerful).
Registries & Plugins - Autopilot now supports users writing their code outside of the library as plugins! To support this, a registry system was implemented throughout the program. Plugin objects can be developed as objects that inherit from the Autopilot object tree – eg. implementing a GPIO object by subclassing
hardware.gpio.GPIO, or a new task by subclassing
Task. This system is flexible enough to allow any lineage of objects to be included as a plugin – stimuli, tasks, and so on – and we will be working to expand registries to every object in Autopilot, including the ability for plugins to replace core modules to make Autopilot’s flexibility verge on ludicrous. The basic syntax of the registry system is simple and doesn’t require any additional logic beyond inheritance to be implemented on plugin objects –
autopilot.get('object_type', 'object_name')is the basic method, with a few aliases for specific object types like
autopilot.get_hardware(). Also thanks to Arne Meyer for submitting an early draft of the registry system and Mikkel Roald-Arbøl for raising the issue.
At long last, the Autopilot Wiki is alive!!!! - https://wiki.auto-pi-lot.com/ - The wiki is the place for communal preservation of technical knowledge about using Autopilot, like hardware designs, build guides, parameter sets, and beyond! This isn’t any ordinary wiki, though, we got ourselves a semantic wiki which augments traditional wikis with a rich system of human and computer-readable linked attributes: a particular type of page will have some set of attributes, like a page about a 3D printed part will have an associated .stl file, but rather than having these be in plaintext they are specified in a format that is queryable, extensible, and infinitely mutable. The vision for the wiki is much grander (but not speculative! very concrete!) than just a place to take notes, but is intended to blend the use of Autopilot as an experimental tool with body of knowledge that supports it. Autopilot can query the wiki with the
wiki.ask('[[Category:3D_CAD]]', 'Has STL')to get links to all .stl files for all 3D parts on the wiki. The integration between the two makes using and submitting information trivial, but also makes designing whole new types of community interfaces completely trivial. As a first pass, the Wiki will be the place to index plugins, the system for submitting them, querying them, and downloading them only took a few hours and few dozen lines of code to implement. The wiki is infinitely malleable – that’s the point – and I am very excited to see how people use it.
Tests & Continuous Integration with Travis! We are on the board with having nonzero tests! The travis page is here: https://travis-ci.com/github/wehr-lab/autopilot and the coveralls page is here: https://coveralls.io/github/wehr-lab/autopilot . At the moment we have a whopping 27% coverage, but as we build out our testing suite we hope that it will become much easier for people to contribute to Autopilot and be confident that it works!
- New Hardware Objects
cameras.PiCamera- A fast interface to the PiCamera, wrapping the picamera library, and using tips from its developer to juice every bit of speed i could!
The I2C_9DOF object was massively improved to take better advantage of its onboard DSP and expose more of its i2c commands.
- New Transforms
timeseries.Kalman- adapted a Kalman filter from the wonderful filterpy package! it’s in the new timeseries transform module
geometry.IMU_Orientation- IMU_Orientation performs a sensor fusion algorithm with the Kalman Filter class to combine gyroscope and accelerometer measurements into a better estimate of earth-centric roll and pitch. This is used by the IMU class, but is made independent so it can be used without an Autopilot hardware object/post-facto/etc.
timeseries.Filter_IIR- Filter_IIR implements scipy’s IIR filter as a transform object.
timeseries.Integrate- Integrate adds successive numbers together (scaled by dt if requested). not much by itself, but when used with a kalman filter very useful :)
geometry.Rotate- use scipy to rotate a vector by some angle in x, y, and/or z
geometry.Spheroid- fit and transform 3d coordinates according to some spheroid - used in the IMU’s accelerometer calibration method: given some target spheroid, and some deformed spheroid (eg. a miscalibrated accelerometer might have the x, y, or z axis scaled or offset) either explicitly set or estimated from a series of point measurements, transform future input given that transformation to correct for the deformed source spheroid.
- New Prefs
'AUTOPLUGIN'- Attempt to import the contents of the plugin directory,
'PLUGIN_DB'- filename to use for the .json plugin_db that keeps track of installed plugins’,
'PING_INTERVAL'- How many seconds should pilots wait in between pinging the Terminal?’,
'TERMINAL_SETTINGS_FN'- filename to store QSettings file for Terminal’,
'TERMINAL_WINSIZE_BEHAVIOR'- Strategy for resizing terminal window on opening’,
'TERMINAL_CUSTOM_SIZE'- Custom size for window, specified as [px from left, px from top, width, height]’,
Multihop messages & direct messaging - https://github.com/wehr-lab/autopilot/pull/99 - it is now possible to send multihop messages through multiple Station objects, as well as easier to send messages directly between net nodes. See the examples in the network tests section of the docs.
Multiple Children (Thank you Chris Rodgers!) - https://github.com/wehr-lab/autopilot/pull/103 - the
CHILDIDfield now accepts a list, allowing a Pilot to initialize child tasks on multiple children. (this syntax and the hierarchical nature of pilots and children will be deprecated as we refactor the networking modules into a general mesh system, but this is lovely to have for now :)
Programmatic Setup - https://github.com/wehr-lab/autopilot/issues/33 - noninteractive setup of prefs and scripts by using
autopilot.setup -f prefs.json -p PREFNAME=VALUE -s scriptname1 -s scriptname2
Widget to stream video, en route to more widgets for direct GUI control of hardware objects connected to pilots
Support python 3.8 and 3.9 essentially by not insisting that the spinnaker SDK be installable by all users (which at the time was only available for 3.7)
Pilots will periodically ping the Terminal again, Terminal can manually ping Pilots that may have gone silent - https://github.com/wehr-lab/autopilot/pull/91
Pilots share their prefs with the Terminal in their initial handshake - https://github.com/wehr-lab/autopilot/pull/91
Reintroduce router ports for net-nodes to allow them to bind a port to receive messages - https://github.com/wehr-lab/autopilot/pull/115/commits/35be5d634d98a7983ec3d3d6c5b94da6965a2579
Listen methods are now optional for net_nodes
Allowed the creation of dataless tasks - https://github.com/wehr-lab/autopilot/pull/115/commits/628e1fb9c8fcd15399b19b351fed87e4826bc9ab
Allowed the creation of plotless tasks - https://github.com/wehr-lab/autopilot/pull/115/commits/08d99d55a32b45f54e3853813c7c71ea230b25dc
I2C_9DOFclas uses memoryviews rather than buffers for a small performance boost - https://github.com/wehr-lab/autopilot/pull/115/commits/890f2c500df8010b50d61f64e2755cd2c7a8aeed
Phasing out using
Queues in favor of
collections.dequefor applications that only need thread and not process safety because they are way faster and what we wanted in the first place anyway.
New Scripts -
utils.NumpyEncoder and decoder to allow numpy arrays to be json serialized
calibrations are now loaded by hardware objects themselves instead of the extraordinarily convoluted system in
prefs– though some zombie code still remains there.
Net nodes know their ip now, but this is a lateral improvement pending a reworking of the networking modules.
performancescript now sets
swappiness = 10to discourage the use of swapfiles - see https://www.raspberrypi.org/forums/viewtopic.php?t=198765
Setting a string in the
deprecationfield of a pref in
_DEFAULTSprints it as a warning to start actually deprecating responsibly.
Logging in more places like Subject creation, manipulation, protocol assignation.
Loggers would only work for the last object that was instantiated, which was really embarassing. fixed - https://github.com/wehr-lab/autopilot/pull/91
Graduation criteria were calculated incorrectly when subjects were demoted in stages of a protocol - https://github.com/wehr-lab/autopilot/pull/91
LED_RGB ignores zero - https://github.com/wehr-lab/autopilot/pull/98
Fix batch assignment window crashing when there are subjects that are unassigned to a task - https://github.com/wehr-lab/autopilot/pull/115/commits/e42fc5802792822ff5a53a2379041a4a8b301e9e
Catch malformed protocols in batch assignment widget - https://github.com/wehr-lab/autopilot/pull/115/commits/2cc8508a4bf3a6d49512197dc72433c60d0c656e
Terminal.reset_uimethod and made control panel better at adding/removing pilots - https://github.com/wehr-lab/autopilot/pull/91
Subject class handles unexpected state a lot better (eg. no task assigned, no step assigned, tasks with no data.) but is still an absolute travesty that needs to be refactored badly.
The jackclient would crash with long-running continuous sounds as the thread feeding it samples eventually hiccuped. Made more robust by having jackclient store samples locally int he sound server rather than being continuously streamed from the queue.
PySide2 references still incorrectly used
pigpio scripts would not be stopped and removed when a task was stopped, the
gpio.clear_scripts()function now handles that.
xcbwas removed from
PySide2distributions, so it’s now listed in the requirements for the Terminal and made available in the
LED_RGBdidn’t appropriately raise a
ValueErrorwhen called with a single
A fistful of lingering Python 2 artifacts
continuing to split out modules in
autopilot.core- networking this time
utils is now a separate module instead of being in multiple places
the npyscreen forms in
setup_autopilotwere moved to a separate module
setup_autopilotwas broken into functions instead of a very long and impenetrable script. still a bit of cleaning to do there.
autopilot.setup.setup_autopilotwas always extremely awkward, so it’s now been aliased as
the docs have now been split into subfolders rather than period separated names to make urls nicer – eg /dev/hardware/cameras.htm rather than /dev/hardware.cameras.html . this should break some links when switching between versions on readthedocs but other than that be nondestructive.
new Quickstart documentation with lots of quick examples!
check_compatiblemethod in the Transforms class. We will want to make a call at some point if we want to implement a full realtime pipelining framework or if we want to use something like luigi or joblib or etc. for now this is an admission that type and shape checking was never really implemented but it does raise some exceptions sometimes.