e.g. don't write methods like
— Jake VanderPlas (@jakevdp) October 13, 2019
def load_data(filename):
data = json.load(filename)
# ...
This means *you* are responsible to support every imaginable filetype/database/etc. Instead write methods like
def load_data(data_dict):
# ...
Let your users use the language.
"Stabilizing Generative Adversarial Network Training: A Survey" -- imho, training GANs can be extremely frustrating; this is a nice paper to keep handy for these occasions: https://t.co/OsbnxXVhb4
— Sebastian Raschka (@rasbt) October 3, 2019
I do the opposite and also tell my students to. I find by keeping the randomness as I experiment I get a better intuitive feel for the modeling problem.
— Jeremy Howard (@jeremyphoward) September 20, 2019
Very interesting to hear your very different approach! :)
- Your model runs will have the exact same perplexity spikes (hits confusing data at the same time)
— Smerity (@Smerity) September 19, 2019
- You can compare timestamp / batch results in early training as a pseudo-estimate of convergence
- Improved gradient flow visibly helps the same init do better
Deep learning training tip that I realized I do but never learned from anyone - when tweaking your model for improving gradient flow / speed to converge, keep the exact same random seed (hyperparameters and weight initializations) and only modify the model interactions.
— Smerity (@Smerity) September 19, 2019
DID YOU KNOW that @jimhester_'s excellent lintr package works really well on Kaggle? If you want your R code looking 💯💯 you should definitely check it out. 💻👀
— Rachael Tatman (@rctatman) September 19, 2019
Sample code:https://t.co/YdpKoRermL
Did you know that Dask can hand off parallel datasets to XGBoost for distributed training?
— Dask (@dask_dev) September 19, 2019
Today there are two implementations, one from Dask, and one from XGBoosthttps://t.co/OJuS1f8yjshttps://t.co/BtEyY5NQq1
SQL tricks: single quote your strings pic.twitter.com/HTbgHLNXkb
— 🔎Julia Evans🔍 (@b0rk) September 17, 2019
Some more🔥for y'all:
— Tim Vieira (@xtimv) September 16, 2019
🔥Now 10x speedup over numpy w/ the "Exp-sort trick" (the cooler cousin of the "Gumbel-max trick")
🔥Exp-sort trick is trivial to parallelize/vectorized
🔥Give a Gumbel-sort trick for log weights
🔥1-pass algs that work w/streamshttps://t.co/su0ufYS67h https://t.co/X39gGfa48w
Dask isn't only about Arrays and Dataframes.
— Dask (@dask_dev) September 16, 2019
Many people use Dask Futures to parallelize existing workloads, and build advanced real-time systems easily that adapt to changing situations.https://t.co/FU30drj8UN pic.twitter.com/Ml0fd2Tee2
It would be reasonable to error out at `x + y` due to unclear broadcasting expectations. It would also be reasonable to broadcast the (32,) vector to (32, 1).
— François Chollet (@fchollet) September 15, 2019
But how about a third, insane option?
Advice for effective analytical thinking. https://t.co/57k5wGS7zI pic.twitter.com/kQfDK326Or
— Edward Tufte (@EdwardTufte) September 11, 2019
