Playground Rasa on a Raspberry Pi connected to Alexa

ubuntu 20.04 / python 3.8 / tensorflow 2.3 / rasa 2.3

It's only a test project to get RASA running on a raspberry pi and connect to an Alexa Skill. It works also with win 10 or macos

Install a raspberry pi

I didn't found all needed packages for the raspberry pi to get rasa running. That's why I build tensorflow with text and addons myself.

• base image: ubuntu-20.04.2-preinstalled-server-arm64+raspi.img
• I used a raspberry pi 4 with 8 GB ram. This works fine

some usefull settings

• sudo dpkg-reconfigure keyboard-configuration
• sudo ufw allow ssh

install libs

• sudo apt install libffi-dev libbz2-dev liblzma-dev libsqlite3-dev libncurses5-dev libgdbm-dev zlib1g-dev libreadline-dev libssl-dev tk-dev build-essential libncursesw5-dev libc6-dev openssl git
• sudo apt-get install libopenblas-base libblas-dev liblapack-dev cython libatlas-base-dev openmpi-bin libopenmpi-dev python3-dev
• sudo apt-get install pkg-config libhdf5-dev
• sudo apt-get install libpq-dev

create python venv

• sudo apt-get install python3-venv
• python3 -m venv /home/ubuntu/rasa1/venv
• source ./rasa1/venv/bin/activate

build and install tensorflow

install bazel

• see https://github.com/PINTO0309/Bazel_bin
• https://github.com/PINTO0309/Bazel_bin/blob/main/3.7.2/aarch64/install.sh

build tensorflow 2.3 (venv activated)

• the build takes a long time on the raspberry (>10 hours)
• see also https://www.tensorflow.org/install/source
• git clone https://github.com/tensorflow/tensorflow.git
• cd tensorflow
• git checkout r2.3
• ./configure
• bazel build -c opt --local_ram_resources=4096 --local_cpu_resources=HOST_CPUS*.5 --config=noaws --copt="-funsafe-math-optimizations" --copt="-ftree-vectorize" --copt="-fomit-frame-pointer" --verbose_failures tensorflow/tools/pip_package:build_pip_package
• so not use: not --copt="-mfpu=neon-vfpv4"
• bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
• copy created wheel to folder of choice
• pip3 uninstall tensorflow
• pip3 install tensorflow-2.3.1-cp38-cp38-linux_aarch64.whl
• ? pip3 install keras_applications==1.0.8 --no-deps
• ? pip3 install keras_preprocessing==1.1.0 --no-deps
• ? pip3 install h5py==2.9.0
• ? pip3 install pybind11
• ? pip3 install -U ser six wheel mock

##build tensorflow-text (venv activated)

• see also https://github.com/tensorflow/text
• git clone https://github.com/tensorflow/text.git
• cd text
• git checkout 2.3
• ./oss_scripts/run_build.sh
• copy created wheel to folder of choice
• pip3 install tensorflow_text-2.3.0-cp38-cp38-linux_aarch64.whl

build tensorflow addons (venv activated)

• git clone https://github.com/tensorflow/addons.git
• cd addons/
• git checkout r0.11
• python3 ./configure.py
• bazel build build_pip_pkg
• bazel-bin/build_pip_pkg artifacts
• copy created wheel to folder of choice
• pip3 install tensorflow_addons-0.11.2-cp38-cp38-linux_aarch64.whl

install rasa

• pip3 install rasa==2.3.4
• You start rasa with custom python script if rasa returns errors with scikit-learn.
  • insert in rasa start-script (in /venv/bin/): import sklearn

install rasa x

• pip3 install rasa-x --extra-index-url https://pypi.rasa.com/simple
• if problems on mac you could try: pip install --upgrade pip==20.2
• start with rasa x (will start rasa x on port 5002 and rasa server on port 5005 but not actions server)
• windows only: if rasa x starts with errors encoding and utf-8 _> set PYTHONUTF8=1

create sqlite db to log coversations

• can be created but rasa will create one if not there when starting server (e.g. in folder of bot: sqlite3 bot1.sqlite)
• add entry in endpoints.yml

install ngrok if you want to connect to alexa skill

• sudo snap install ngrok
• If you create a free account you can get a token and will not have the 2 hour limitation

connect to Alexa

• see the very good description and video https://blog.rasa.com/connect-your-rasa-ai-assistant-to-amazon-alexa/
• POST from alexa go to http://server:5005/webhooks/alexa_assistant/webhook
• check status with http://server::5005/webhooks/alexa_assistant

spacy and other language

• pip install spacy
• python -m spacy download de_dep_news_trf
• python -m spacy download de_core_news_sm

proxy (windows)

• rasa action server could have problems to connect rasa x with activated proxy set http_proxy= set https_proxy=

install duckling for DucklingEntityExtractor

• sudo apt install haskell-stack
• sudo apt install ghc
• sudo apt install libpcre3 libpcre3-dev
• git@github.com:facebook/duckling.git
• cd duckling
• stack build --system-ghc
• start with : stack exec duckling-example-exe --system-ghc

python

• freeze current installtion settings e.g. : python -m pip freeze >requirements_rasa25.txt

ports

• rasa server : 5005 ( check with http://server:5005/version )
• action server : 5055
• rasa x : 5002 (default with password. see when rasa x has started)
• duckling : 8000