develocorder 0.3.0

Simple live value plotter using Matplotlib

Develocorder

Develocorder is a simple live value plotter for Python3 using Matplotlib. It is intended to give insights into the training of machine learning models / reinforcement learning agents with only minimal effort to add to existing code.

(image source)

Installation

$ pip install develocorder

Simple Example

# initialize once
set_recorder(my_value=LinePlot())

# add values to plot from anywhere in code
for _ in range(10):
    record(my_value=random())

See examples/simple.py

Result

Fancy Example

Some more features:

# axis labels
set_recorder(score=LinePlot(xlabel="Episode", ylabel="Score"))

# filter values (window filter kernel)
set_recorder(loss=LinePlot(filter_size=64))

# maximum history length
set_recorder(loss_detail=LinePlot(max_length=50))

# show heatmap for recording 1d-array values
set_recorder(array_values=Heatmap(max_length=1000))

# minimum update period (limit rate a which graphs are redrawn for better performance)
set_update_period(0.5)  # [seconds]

# set number of columns
set_num_columns(2)

See examples/fancy.py

Jupyter notebook

For use in a Jupyter notebook use the %matplotlib notebook backend. As of now you cannot rerun the cell which is showing the plot without restarting the notebook, otherwise the plot will disappear.

See examples/simple.ipynb

TODOs

• document how to extend
• better support for jupyter notebook
• add new plot types
• persistent storage/loading of log