Power Utility Tools for DNN analysis
Gathering data via predefined test cases in console (see script for available options):
python3 ./tests/test_measure.py
Instantiation in code:
from powerutils import measurement
# create instance of class
pm = measurement.power_measurement(
sampling_rate=500000, # set the sampling rate to whatever your device supports
data_dir = "./tmp", # pass the folder where data files will be saved
max_duration=60, # set the maximal duration of the gathering process [seconds]
port=0, # if your DAQ device has more than one port, choose the port where the DUT is connected to
range_index=3) # choose a measurement resolution from available ranges (the higher, the more accurate)
# define parameters for the name of the data file
test_kwargs = {"model_name" : "awesome_model", "index_run" : 1, "my_parameter" : "some_value"}
pm.start_gather(test_kwargs) # start the data aquisition
# here should be the inference on a platform
from time import sleep; sleep(2); # or a sleep command to test the data gathering
pm.end_gather(True) # ends the data gathering and writes it to a data (.dat) file
print("Finished")
Clone the repository to your machine and navigate into it:
git clone https://github.com/embedded-machine-learning/powerutils.git cd powerutils
(OPTIONAL) Create a Python3 virtual environment and activate it:
python3 -m venv venv_powerutils source venv_powerutils/bin/activate
Install powerutils locally and check the installation:
pip3 install -e . python3 -c "import powerutils; help(powerutils)"
The last command should show general information of the module. Exit help() by typing "q" (without the quotation marks)
Linux machine with Python3 installed (tested on Ubuntu 18.04 LTS) A Data Aqcuisition Card from https://www.mccdaq.com Python3 modules: uldaq, numpy, pandas, matplotlib OpenVino for the profiling of the Intel Neural Compute Stick 2 TF Lite for profiling of the Google Edge TPU
powerutils was written by CDL EML.