tudor lapusan

Big Data/ML enthusiast.

[Terence Parr @the_antlr_guy]

New in https://t.co/bUyeA0CBle: AI-powered tree explanations!
Chat with decision trees with natural language. Ask about tree structure, node splits, feature importance, leaf predictions, etc. Get detailed insights powered by OpenAI.
Enable with ai_chat=True. Thanks to @tlapusan!

[Terence Parr @the_antlr_guy]

A blog entry I wrote for google/tensorflow on using dtreeviz for decision tree visualization. :)

[Terence Parr @the_antlr_guy]

Blog post I wrote for TensorFlow on integration of https://t.co/bUyeA0CBle viz library for decision forests :) Work done with @tlapusan

[TensorFlow @TensorFlow]

Decision Forests have an important advantage: Explainability 📝

The dtreeviz library takes this to the next level with powerful, helpful, and beautiful visualizations!

Learn how to use it with TFDF and more in this blog post ➡️ https://t.co/ENdtCwVLaT

[Terence Parr @the_antlr_guy]

Check out the tutorial on using dtreeviz for visualizing tensorflow decision forests! https://t.co/7s6ioeHCg2

[Terence Parr @the_antlr_guy]

Released dtreeviz 2.1 cleanup release (code and visualizations). E.g., leaves can be bars not just pies now. With Matthew Epland and @tlapusan. Notes: https://t.co/3jZ6DXFgtB

[Terence Parr @the_antlr_guy]

Released 2.0.0 dtreeviz! Re-organizes the API to focus on using a model adaptor vs straight function calls; should be backward compatible though. Supports scikit-learn, XGBoost, Spark MLlib, LightGBM, Tensorflow. A pleasure working w/coauthor @tlapusan.
https://t.co/0N3CHLp7KE

[Sebastian Raschka @rasbt]

Looking for ways to spruce up my decision tree lecture this week & stumbled upon the dtreeviz library compatible with scikit-learn: https://t.co/lxeFfx9pUa. Not sure if I love the pie charts, but the histograms plus split indicator are kind of nice. Also nice viz for regr trees

[tudor lapusan @tlapusan]

@rasbt @jwuphysics Also the library contains visualisations for tree leaves details, like number of samples and impurities, viz_leaf_criterion() and viz_leaf_samples(). In the end the tree performance is reflected by the performance of each leaf.

[tudor lapusan @tlapusan]

@rasbt @jwuphysics dtreeviz() method contains the depth_range_to_display parameter which allows you to specify a range of tree levels to display only.

[Jeremy Howard @jeremyphoward]

But once random forests appeared on the scene, I started using them pretty much everywhere!

(Diagram from dtreeviz by @the_antlr_guy)

[Terence Parr @the_antlr_guy]

Added classifier prediction probability visualization for 1D/2D feature space; works for any model answering predict_proba() (e.g., sklearn, keras,...). Great for edu! Resides in dtreeviz to avoid yet another package: `pip install -U dtreeviz` to get 1.3. https://t.co/LCQp92wjYc

[tudor lapusan @tlapusan]

@BurakSayici @the_antlr_guy @AndreasStefik indeed, big trees are hard to interpret by looking at their structure (regardless of the visualisation format). When I'm dealing with big trees, the most important information are from the leaves, like impurity and number of samples. dtreeviz contains visualisations for them also

[Terence Parr @the_antlr_guy]

Pushed dtreeviz (decision tree visualizer) 1.2 with the support for #LightGBM, thanks to @tlapusan; `pip install -U dtreeviz`

[Oege Dijk @oegedijk]

@the_antlr_guy @XGBoostProject @tlapusan Added xgboost dtreeviz support to explainerdashboard! Also added a visualization of the individual trees marginal contribution: https://t.co/xz9ZasWjCh

[Terence Parr @the_antlr_guy]

dtreeviz 1.1 now visualizes @ApacheSpark decision trees (thanks to @tlapusan) in addition to @XGBoostProject and @scikit_learn. `pip install -U dtreeviz` or see https://t.co/bUyeA0TEne

[Terence Parr @the_antlr_guy]

Added @XGBoostProject support for decision tree visualization in dtreeviz 1.0 thanks to @tlapusan. :) Get it while it's hot: `pip install -U dtreeviz`. See https://t.co/bUyeA0TEne #visualization #MachineLearning

[Valentina Crisan @ValentinaCrisan]

Machine learning intro workshop on July 13th, Python and SKlearn, @tlapusan trainer and using @lentiq to run and share our notebooks. Looking forward to it. https://t.co/RWD0i7bMfK

[tudor lapusan @tlapusan]

@SparkAISummit Do you have a date for spark summit europe 2018?

[tudor lapusan @tlapusan]

@myadav2009 @kdnuggets the labels are correct. You can check this video about bias vs variance from @AndrewYNg https://t.co/dwMkGQtJ5M