Jorge Torres - Machine Learning Democratization with Python
1 like88 views
This document discusses democratizing machine learning by empowering developers to utilize ML through simple tools. It notes that true ML experts are rare and in high demand. The approach is to abstract away complexity and expose ML via tools developers already know how to use. This will allow more people beyond just ML experts to exploit ML. Goals include making ML frictionless to use with just a line of code, ensuring explanations of predictions, and staying on the cutting edge of ML advances. The upcoming MindsDB version 1.0 is outlined as having a cleaner API, support for more data types, options for TensorFlow or PyTorch, enhanced explainable AI for model and prediction quality, and capabilities for production systems.
![Why Democratizing M.L?
Note: True ML experts are rare and on high demand [1].
[1] https://www-01.ibm.com/common/ssi/cgi-bin/ssialias?htmlfid=IML14576USEN&
Demand for
predictions ->
Î
ML Expert
-> predictions
& insights](https://image.slidesharecdn.com/tsb0isxrxg4fzqyokmt7-signature-d54a379aa08ffdc2f6d36f4673758e047c55b447669585938c3e826a2f57439b-poli-190321205606/85/Jorge-Torres-Machine-Learning-Democratization-with-Python-3-320.jpg)
![> More Data types
From:
------
- NUMERIC
- CATEGORICAL
To:
-----
- NUMERIC
- INT
- FLOAT
- CATEGORICAL
- SINGLE
- MULTIPLE
- DATETIME
- STRING
- TIMESTAMP
- TIMESTAMP_DELTA
- BINARY
- IMAGE
- AUDIO
- VIDEO
- SEQUENTIAL
- TEXT
- NESTED l
- NUMERIC
- DATETIME
- PAIRS
# example pairs of [[NUMERIC, TIMESTAMP_DELTA], ….]](https://image.slidesharecdn.com/tsb0isxrxg4fzqyokmt7-signature-d54a379aa08ffdc2f6d36f4673758e047c55b447669585938c3e826a2f57439b-poli-190321205606/85/Jorge-Torres-Machine-Learning-Democratization-with-Python-16-320.jpg)
Jorge Torres - Machine Learning Democratization with Python
- 2. Why M.L? - Use the data that we have to make informed predictions.
- 3. Why Democratizing M.L? Note: True ML experts are rare and on high demand [1]. [1] https://www-01.ibm.com/common/ssi/cgi-bin/ssialias?htmlfid=IML14576USEN& Demand for predictions -> Î ML Expert -> predictions & insights
- 4. How to Democratize M.L? - Simple ways for other domain experts (non ML people) to exploit ML on their own. Demand for predictions -> -> predictions & insights
- 5. First step? - Empower Developers that are NOT ML Experts. Why? - Developers are AWESOME! we build shit, we get it done! - Developers have easier access to data and means to aggregate it. - There are many more developers than ML experts out there.
- 6. How? - Abstract complexity and expose it via tools developers already know.
- 7. What’s important - X.A.I: Explainable AI. Why? - If we are delegating the ML complexity to a system, we need to trust the system.
- 9. Goal 1: Frictionless - Be able to create and use complex predictive models with no more than a line of code.
- 10. Goal 2: X.A.I Ready Must answer: - Why this prediction? - Why not something else? - When is the model working? - When is the model not working? - When can I trust the model? - When can’t I trust the model? - How can I make my models better
- 11. Goal 3: Stay State Of Art - ML/AI moves fast so do we.
- 12. pip3 install mindsdb Thank you! :)
- 13. pip3 install mindsdb Just kidding, not done yet
- 14. What is coming this month? MindsDB V.1.0 - Cleaner API - Many more data types - Tensorflow/Pytorch you pick - XAI (Quality domain) - Data quality - Model quality - Prediction quality - Production system
- 15. > Cleaner API Introducing Predictor objects
- 16. > More Data types From: ------ - NUMERIC - CATEGORICAL To: ----- - NUMERIC - INT - FLOAT - CATEGORICAL - SINGLE - MULTIPLE - DATETIME - STRING - TIMESTAMP - TIMESTAMP_DELTA - BINARY - IMAGE - AUDIO - VIDEO - SEQUENTIAL - TEXT - NESTED l - NUMERIC - DATETIME - PAIRS # example pairs of [[NUMERIC, TIMESTAMP_DELTA], ….]
- 17. > XAI From: ------ - None :( To: ----- predictor.explain_quality() prediction.explain_quality()
- 18. > XAI predictor.explain_quality() - DATA QUALITY - Evaluate quality of the data that you train from AND tell you about what we find: - Consistency: How much error - Redundancy: Repeated data in columns and rows - Variability: - Biases - Noise/outliers - No variation - Completeness: How much missing data - MODEL QUALITY - What values its good at predicting - What values its not good at predicting - What data quality issues are responsible for this
- 19. > XAI prediction.explain_quality() - Predicted value - Certainty of predicted value - Probability distribution of possible values
- 20. > From desktop to production
- 21. Thank you This time for real, bye ;) Ah! And ask me about early access to all this