In this work, I will be building several functions for the Furby in different neural networks. The first one I am working on in Python, is the function of speech, response. The neural networks will work through different microcontrollers to test and respond to before integrated into the Furbies robotic form. That is currently completely stripped. To be fair this is a Chebbacca Furby.
Personally I love Star Wars and I love Furbies!
Raspberry Pi recently came out with a Pi Zero 2 W. The board specs are phenominal and you can read those here. https://www.raspberrypi.com/products/raspberry-pi-zero-2-w/ This is the board I will use to begin working on the Furby's brain. I call it a brain, but in all reality it is a AI I am developing to integrate into the Furby.
Learning Distribution Symantics is no easy task with creating the first neural network for speech and response. Distribution Symantic modeling is an interesting take on words. You can find details of Distribution Symantic Modeling on Wikipedia https://en.wikipedia.org/wiki/Distributional_semantics .
Its a entirely different approach than the sensor of photo recognition. There are many scholar writings I have been combing through such as this one "Distributional semantic modeling: a revised technique to train term/word vector space models applying the ontology-related approach." by Vitalii Velychko
https://www.researchgate.net/publication/339713296_Distributional_semantic_modeling_a_revised_technique_to_train_termword_vector_space_models_applying_the_ontology-related_approach
He explores a reasonably interesting Library in Python that I will begin working with, Vec2graph -- a Python library for visualizing word embeddings (term embeddings in my case) as dynamic and interactive graphs. This is also being explored by Cornell University and a few other scholars. It is a good read on how words explored. I also have the PDF loaded into the "Furby Resource" Folder above if you want to have easy access to that information.
Considering I have never done this before, I am going to start following this approach as a base set up. Part one says to create a Dataset using a large wikipedia data dump set, with Python Gensim to parallelize and process the dump. I need storage off site from my computer because that dump is huge.
((Hardware needed))
GEFORCERTX 4060 Ti (its important to ensure you have atleast 16VRAM)
I have been working on configuring these hardware specs before I begin working with the wikipedia dump, as to ensure the tower I have can handle the load.
This would be minimium specs required for this project to get started, I am sure this will still be slightly slower, but once configured, optimizations will be made. This is a process I hope no one minds following along with me. I will gladly share the steps as I walk threw them and the code to get this going. Sorry I haven't worked on this in a while...personal things came up-
Ok Hardware is set :)
**Commmands on Linux to use to check GPU setup:
nvidia-smi
nvidia-smi dmon
sudo apt install nvtop
nvtop
I am back at it now-
Next is configuring Apache Spark on Ubuntu for handling the xml files which are huge.
PLS - don't use ChatGPT [[warning it will not work no matter what it tells you to do]]
https://www.virtono.com/community/tutorial-how-to/how-to-install-apache-spark-on-ubuntu-22-04-and-centos/
-us that site it works.
Wiki-dump site -
https://dumps.wikimedia.org/enwiki/20240701/
enwiki-20240701-pages-meta-history1.xml-p1p812.bz2 2.4 GB
enwiki-20240701-pages-meta-history1.xml-p813p1460.bz2 2.0 GB
This is kind of hard to configure so I had to make sure that my tower specs were set properly to handle this much data.
Total Memory: 62 GiB
Used Memory: 4.2 GiB
Free Memory: 40 GiB
Total Swap: 15 GiB
Used Swap: 0 B
Bash: watch free -h
Reached an big error with schemas in the xml chunk files. https://www.youtube.com/watch?v=aB_koPUNqfo this Youtube video helped with debugging the schemas of the namespace issues in building a element tree.