András Gyácsok's blog

My impressions about the presentation. I watched it with strong motivation to find a way for my project, a recommendation process for a chain of process unit steps. A recommendation is computed on a combination of ML models, which are dynamically created during the time. I guess service choreography is good way to go for me. I like conception of no contract, no implementation aware solutions. Everything is transmitted in the system through events. However, I am not sure is it possible to implement such an architecture with Python. Will keep searching. Also, interesting point is in the building communication schema using stickers of different colors.

https://docs.vespa.ai/documentation/vespa-quick-start.html https://github.com/vespa-engine/vespa/tree/master/storageapi/src/vespa/storageapi/message

There is a way to serve multiple models within one Tensorflow Serve server providing config file. model_config_list: { config: { name: "model1", base_path: "/models/model1", model_platform: "tensorflow" }, config: { name: "model2", base_path: "/models/model2", model_platform: "tensorflow" } } More detailed information possible to get here . What I would like to know how convenient is this solution? Is it possible to upload new versions? Edit config?

https://medium.com/dartlang/darts-built-value-for-immutable-object-models-83e2497922d4#.48dyezxcl https://medium.com/dartlang/darts-built-collection-for-immutable-collections-db662f705eff https://medium.com/dartlang/darts-built-value-for-serialization-f5db9d0f4159#.h12y94wu7

Found super interesting article about speaker diarization. Guys build full framework to separate speakers on audio online. The paper can be found here . The repository is here . Other interesting articles on topic: https://www.infoq.com/news/2018/11/Google-AI-Voice/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5996252/ http://theconversation.com/human-voices-are-unique-but-our-study-shows-were-not-that-good-at-recognising-them-79520 https://towardsdatascience.com/automatic-speaker-recognition-using-transfer-learning-6fab63e34e74 https://medium.com/@ageitgey/machine-learning-is-fun-part-6-how-to-do-speech-recognition-with-deep-learning-28293c162f7a https://medium.com/linagoralabs/voice-activity-detection-for-voice-user-interface-2d4bb5600ee3 https://medium.com/linagoralabs/computing-mfccs-voice-recognition-features-on-arm-systems-dae45f016eb6 https://towardsdatascience.com/beginners-guide-to-speech-analysis-4690ca7a7c05 https://towardsdatascience.com/speech-recognition-i...

From my point of view very nice course for free. I highly appreciate the work of Nick. Clean code practises in course are really good.

We found this receipt today with my middle daughter. Waffles were so delicious :)

Interesting ideas found here  and here . Cool solution  youtube-dl .

Want to get to know Flutter well - here is GIthub repository with the application to learn.

Spinx Open-source application for voice-to-text processing. It is written on C++ and Java, Python client exists. Code can be found here . Project page .

LSTM Attention

Here is a list of sites, helping estimate vulnerabilities  of own site: https://report-uri.com https://securityheaders.com/

https://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/ https://www.youtube.com/watch?v=d14TUNcbn1k&t=2273s

Here below provided a bunch of articles about speech processing and noise suppression:   Audio features for web-based ML  - nice article about in-browser speech recognition.   Simple Audio Recognition  - article about Tensorflow usage for audio classification.   Real-Time Noise Suppression Using Deep Learning  - article about noise suppression   Denoise Speech Using Deep Learning Networks  - Matlab version  RNNoise: Learning Noise Suppression  - Mozilla project (useful) GitHub denoising projects: Wavenet For Speech Denoising Source-Denoising-Pix2Pix-cGAN

For local installation of Kubeflow you can use Vagrant box, Minikube or install it using Multipass. Here I will describe the multipass installation. The best link for the installation process is here . This installation uses Microk8s orchestrator and has some caveats. First of all you have to create volumes manually, other way you services won't work correctly. Before all I would suggest to install version 1.13 of microk8s, because 1.14 is not compatible now with kubflow. sudo snap install microk8s --channel=1.13 For this purpose please edit kubernetes-tools/setup-microk8s.sh line 10: CHANNEL=${CHANNEL:-1.13/stable} 1.14 where changed on 1.13. Then run  sudo kubernetes-tools/setup-microk8s.sh Step 1: Create storage class. I used next data: apiVersion: storage.k8s.io/v1 kind: StorageClass metadata: name: local provisioner: kubernetes.io/no-provisioner volumeBindingMode: WaitForFirstConsumer Just save and apply this data. kubectl apply -f you_file_name.yaml ...

Best article: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/ Best free video: https://www.youtube.com/playlist?list=PL3FW7Lu3i5JvHM8ljYj-zLfQRF3EO8sYv https://www.youtube.com/watch?v=ArPaAX_PhIs&list=PLkDaE6sCZn6Gl29AoE31iwdVwSG-KnDzF Channel https://www.youtube.com/channel/UCcIXc5mJsHVYTZR1maL5l9w

Recently I created data - science model training and upload CI/CD pipeline in OpenShift. I had next idea: Data-scientist creates new model and uploads it in a repository. A repository triggers web-hook and notifies OpenShift about pushed / merged commit. Jenkins starts an execution of commands described in Jenkinsfile. Jenkinsfile in my case contains list of commands to create a container, train a model there, test it, wrap it and then synchronize it  with a responsible Jenkins slave (slave started the process). Then slave uploads an artifact in to a system.  The only problem I met implementing this solution was injecting certificates into a Jenkins slave. I tried different ideas - mounting secrets as volumes to a shared folder in Jenkins master, using Kubernetes plugin form for attaching secrets... - I didn't get any result. Help, I received from my colleague Michael, solution is quite simple, but not easy to found. Create a special type of 'ssh-auth' ...

https://www.youtube.com/watch?v=_HfWDtu7A5I and https://stackoverflow.com/questions/48111084/authentication-by-certificate-for-webflux and  https://blog.codecentric.de/en/2018/08/x-509-client-certificates-with-spring-security/

Solution is well defined in the next article.

Spring boot uses special configuration to set up trailing slashes, which is an overridden method of WebMvcConfigurer - configurePathMatch. More information can be found here .

From my point of view good solution is to use virtualenv there. Below is small snippet of code for a Jenkinsfile. sh """ pip install --user virtualenv && virtualenv venv && . \$(pwd)/venv/bin/activate && pip install -r src/test_requirements.txt && export PYTHONPATH="src" && cd src/tests && nosetests -v """

When you use entities with lazy loading fields you can meet Jaxson error, like " No serializer found....". The best solution and description I have found is here.