Video Title:
AutoML: from data acquisition to predictions in production in a few clicks - Dr. Georgina Tryfou
Contents:
0:00 Waiting
4:45 About PyData Cyprus
8:55 Waiting
9:38 Welcome
10:21 Mission of ML
11:12 What do we offer?
11:54 Customer Journey
15:34 ML Training: The Manual Process
19:34 ML Training: The Automated Process
21:22 The AutoML Library
23:06 AutoML: Preprocessing pipeline
27:12 AutoML: Modelling pipeline
30:15 AutoML: Classification Pipeline configuration
32:20 MLFlow: Introduction & Integration
34:16 MLFlow: Demo
37: 12 AutoML: Results on Titanic Kaggle Dataset
38:02 Problem 1: Intelligent Lead Scoring
38:47 Problem 1: AutoML Results on Lead Scoring problem
41:01 Problem 2: Intelligent Churn Prediction
41:29 Problem 3: AutoML Results on Churn Prediction problem
43:06 Conclusion
43:43 Contact Us
44:49 Q&A 1
47:12 Q&A 2
50:12 Q&A 3
53:36 Q&A 4
Q1. could we use this methodology in combination with NLP to automate a repository to sort article processings etc and fork other repositories for a research community?
Q2. How is it to integrate some new every now and then we see for example pytorch tensorflow coming with new models how easy it is to integrate the model in automl and this is not out of the box it's not a standard model you need a bit of understanding what's going on or with graph neural networks for example thats it's not the typical conventional machine learning models there i guess there is some work need to be done to incorporate these new models?
Q3. In one slide I saw that before training your model you may be doing a dimensionality reduction or standardize your data before training around your model. Is there any chance and despite of gaining more performance scores and better performance scores you may lose interpretability into your result?
Video link: https://www.youtube.com/watch?v=Zth8ZG9q4EY
Video Title: AutoML: from data acquisition to predictions in production in a few clicks - Dr. Georgina Tryfou
Contents:
0:00 Waiting 4:45 About PyData Cyprus 8:55 Waiting 9:38 Welcome 10:21 Mission of ML 11:12 What do we offer? 11:54 Customer Journey 15:34 ML Training: The Manual Process 19:34 ML Training: The Automated Process 21:22 The AutoML Library 23:06 AutoML: Preprocessing pipeline 27:12 AutoML: Modelling pipeline 30:15 AutoML: Classification Pipeline configuration 32:20 MLFlow: Introduction & Integration 34:16 MLFlow: Demo 37: 12 AutoML: Results on Titanic Kaggle Dataset 38:02 Problem 1: Intelligent Lead Scoring 38:47 Problem 1: AutoML Results on Lead Scoring problem 41:01 Problem 2: Intelligent Churn Prediction 41:29 Problem 3: AutoML Results on Churn Prediction problem 43:06 Conclusion 43:43 Contact Us 44:49 Q&A 1 47:12 Q&A 2 50:12 Q&A 3 53:36 Q&A 4
Q1. could we use this methodology in combination with NLP to automate a repository to sort article processings etc and fork other repositories for a research community? Q2. How is it to integrate some new every now and then we see for example pytorch tensorflow coming with new models how easy it is to integrate the model in automl and this is not out of the box it's not a standard model you need a bit of understanding what's going on or with graph neural networks for example thats it's not the typical conventional machine learning models there i guess there is some work need to be done to incorporate these new models? Q3. In one slide I saw that before training your model you may be doing a dimensionality reduction or standardize your data before training around your model. Is there any chance and despite of gaining more performance scores and better performance scores you may lose interpretability into your result?