Install Rasa on Raspbian (Debian) Buster

[chidiewenike]

Objective

Installing a working version of Rasa on Raspbian Buster to be used on a Raspberry Pi.

Key Result

Details

Explore solutions for installing Rasa on Raspbian Buster which requires both Tensorflow and SpaCy.

[chidiewenike]

Preliminary Setup:

sudo apt-get update && sudo apt-get upgrade -y && sudo apt-get dist-upgrade -y

sudo apt-get install libatlas-base-dev python3-dev python3-pip libhdf5-dev python-pip python-dev -y

sudo apt install -y python3-scipy sudo apt install gfortran

sudo reboot

pip3 install rasa-nlu[spacy]

[chidiewenike]

Install TF 2.2 on RPi:

https://qengineering.eu/install-tensorflow-2.2.0-on-raspberry-pi-4.html

get a fresh start

$ sudo apt-get update $ sudo apt-get upgrade

remove old versions, if not placed in a virtual environment (let pip search for them)

$ sudo pip uninstall tensorflow $ sudo pip3 uninstall tensorflow

install the dependencies (if not already onboard)

$ sudo apt-get install gfortran $ sudo apt-get install libhdf5-dev libc-ares-dev libeigen3-dev $ sudo apt-get install libatlas-base-dev libopenblas-dev libblas-dev $ sudo apt-get install liblapack-dev cython $ sudo pip3 install pybind11 $ sudo pip3 install h5py

upgrade setuptools 40.8.0 -> 47.1.1

$ sudo pip3 install --upgrade setuptools

install gdown to download from Google drive

$ pip install gdown

set PATH

$ export PATH=$PATH:/home/pi/.local/bin

download the wheel

$ gdown https://drive.google.com/uc?id=11mujzVaFqa7R1_lB7q0kVPW22Ol51MPg

install TensorFlow

$ sudo -H pip3 install tensorflow-2.2.0-cp37-cp37m-linux_armv7l.whl

and complete the installation by rebooting

$ reboot

[chidiewenike]

Current Issue:

ERROR: Could not find a version that satisfies the requirement tensorflow-addons<0.8.0,>=0.7.1 (from rasa) (from versions: none) ERROR: No matching distribution found for tensorflow-addons<0.8.0,>=0.7.1 (from rasa)

[mfekadu]

https://qengineering.eu/install-tensorflow-2.1.0-on-raspberry-pi-4.html

[mfekadu]

Alternatively run Docker on rpi

https://github.com/rgstephens/rasaPi

• consider Ubuntu on Docker container
• also consider Debian on Docker container

[mfekadu]

Alternatively build from Tensorflow source

• https://www.tensorflow.org/install/source_rpi

[snekiam]

I was able to get this running on my Pi 4 1gb, although I didn't get much further than training a model yet. All I had to do was modify the version of the docker-compose.yml file to 3.3, and it worked after that.

[snekiam]

Overview of how I got Rasa running using this with docker-compose on a Raspberry Pi 4b (1gb):

0. Installed a new version of Docker on my pi by running

   curl -sSL https://get.docker.com | sh 
   sudo usermod -aG docker pi  

   You'll have to re-login (or possibly restart your shell) after doing this.

1. Cloned the above repo to my pi
2. Edited the docker-compose.yml file so that the version is 3.3, and the Rasa version is 1.10.10. The contents of my docker-compose.yml is below:

   version: '3.3'
   services:
   rasa:
   image: rasa:1.10.10
   expose:
     - 5005
   ports:
     - 5005:5005
   volumes:
     - ./:/app
   command:
     - run

3. Changed lines 13 and 14 in endpoints.yml to the following:

   action_endpoint:
   url: "http://app:5055/webhook"

   This sets up the webhook endpoint. I believe app is the hostname inside the docker container.

4. Ran make docker to build the docker image
5. Ran docker-compose run rasa init to setup Rasa inside the image (the appdir will be on persistent storage), using the default directory
6. (Unless I forgot something) You should now be able to run Rasa inside a Docker container by running docker-compose up. If you'd like to run detached, you can run docker-compose up -d.
7. Ask Rasa a test question by ssh'ing into the pi with a separate session, and running:

   curl -XPOST http://localhost:5005/webhooks/rest/webhook \
   -H "Content-type: application/json" \
   -d '{"sender": "test", "message": "Hey"}'

   We probably could run with outside world access, but I'm not entirely sure what that'll take (perhaps something as simple as changing what IP we're listening on from app to 0.0.0.0)

My model was really simple - I didn't change anything from the default. After the initial training, my CPU usage stayed pretty constant below 20% on one core (hovering around 15% most of the time, spiking to 30% only very briefly.) while querying with simple phrases, and my total system ram usage stayed below 400mb. We should try with larger models, but it seems like we may not need a Pi with more ram after all.

[chidiewenike]

@snekiam successfully installed Rasa via Docker. Closing as a result.