roboflex_imgui

Roboflex visualizers and guis using IMGUI/IMPLOT

System dependencies

Requires SDL and GLEW to be installed. More than likely, they already are installed in your distro. But if not:

apt-get install libsdl2-dev
apt-get install libglew-dev

# or maybe on mac:

brew install sdl2
brew install glew
brew install glfw

pip install

pip install roboflex.imgui

Import

import roboflex.imgui as rgu

Nodes

There are two so far: OneDTV and MetricsTelevision

OneDTV (One-Dimensional Television)

It expects to receive messages that contain a tensor under the key <data_key>. That tensor needs to be of shape (C, S), where C is number of channels, and S is sequence length. In general, C should be small (< 20), and S large. Think multiple audio channels.

# all parameters optional: below are the defaults
visualizer = rgu.OneDTV(
    data_key = "data",
    sample_size = 4,
    center_zero = True,
    initial_size = (640, 220),
    initial_pos = (-1, -1),
    name = "OneDTV",
    debug = False,
)

# must be started
visualizer.start()

# NOTE!!!
# On some systems, such as mac, it's bad news to
# run a UI on a non-main thread. So instead of 
# calling start(), you can do this, which will 
# run on the main thread. This call will block 
# until the window is closed, in this case.
visualizer.run()

MetricsTelevision

Real-time visualization of metrics data as published by GraphRoot::start(profile=True). I.e., when you instantiate a GraphRoot (from roboflex core), make it the root of your graph, and call myroot.start(profile=True) on it, it will publish metrics messages. This is the node that can visualize those metrcs message.

# all parameters optional: below are the defaults
visualizer = rgu.MetricsTelevision(
    window_title = "MetricsTelevision",
    initial_size = (1580, 720),
    initial_pos = (-1, -1),
    name = "MetricsTelevision",
    debug = False,
)

# must be started
visualizer.start()

# NOTE!!!
# On some systems, such as mac, it's bad news to
# run a UI on a non-main thread. So instead of 
# calling start(), you can do this, which will 
# run on the main thread. This call will block 
# until the window is closed, in this case.
visualizer.run()

Using MQTT in this case

MQTT is an ideal transport for metrics messages - they are small and infrequent. We provide helper functions in metrics_central.py to do so:

# give one of these to a GraphRoot as it's publisher node
pub = rgu.metrics_central.get_metrics_mqtt_publisher(
    broker_address: str,
    broker_port: int = 1883,
    metrics_topic: str = "metrics",
)

# get a subscriber
sub = rgu.get_metrics_mqtt_subscriber(
    broker_address: str,
    broker_port: int = 1883,
    metrics_topic: str = "metrics",
)

# get a visualizer
vis = rgu.get_metrics_visualizer(
    broker_address: str,
    broker_port: int = 1883,
    metrics_topic: str = "metrics",
)

# connect the subscriber to the visualizer
sub > vis

# Here's a program that does it all - launches a MetricsVisualizer
# listening to some mqtt broker:port and topic.
python3 metrics_central.py