David Kirkby

Recorded seminars: - LBL DL4Sci 2019 school: https://www.youtube.com/playlist?list=PL20S5EeApOSvfvEyhCPOUzU7zkBcR5-eL - LBL DL4Sci 2020 school: - https://dl4sci-school.lbl.gov/agenda - https://www.youtube.com/playlist?list=PL20S5EeApOSuFC7PhHtCWtCsqpNKXM64Q - ML Club: https://docs.google.com/document/u/1/d/1GGtE-YIuAWlmpKSr38_kyiF-Fklszhkh4FkiYWzBAho/pub

Deep double descent: - https://openai.com/blog/deep-double-descent/ - https://arxiv.org/abs/1912.02292

Google ML Crash Course https://developers.google.com/machine-learning/crash-course/ml-intro

I am able to access the CLASS transfer functions within my hacked `ccl_cosmology_compute_power_class()` but I see negative values so I'm not sure how they are defined. Any ideas? In any...

The CLASS transfer functions make sense if I plot `-t(k) / t(kmin) / k**2`, as shown in this plot of the total matter density transfer function compared with BBKS: ...

Yes, I think we still want this. Is it easier to implement now?

Use the following command for profiling: ``` python -m cProfile -o profile.out ./evaluate.py -i word2vec.dat.4 --top-pairs 20 ``` This process runs at 100% CPU (so is single threaded) and uses...

Here are the top-10 entry points: ``` In [1]: import pstats In [2]: p = pstats.Stats('profile.out') In [3]: p.sort_stats('time').print_stats(10) Mon Jan 16 08:16:38 2017 profile.out 197501073 function calls (197470716 primitive...

A tree decomposition of the normalized embedding vectors for all words in the vocabulary should help speed this up a lot. The basic idea is to quickly find a subset...

I just noticed this comment in the KD-tree docs, but I think it is still worth trying this out: > For large dimensions (20 is already large) do not expect...