🔥︎Delighted, thrilled, happy and please to share torchcurves - parametric curves in pure PyTorch. I believe it's mature enough to share here. Inspired by a two-year old paper from Yahoo.
code: https://t.co/MkW8DDKamf
docs: https://t.co/7jQlufbZch
paper that inspired it: https://t.co/FuFbqX8VSI
Use:
pip install torchcurves
or my personal preference:
uv add torchcurves
A small library of vectorized, differentiable B-spline and Legendre curves with learnable coefficients. Handy for continuous embeddings of numerical features in your recommender system, shape-constrained models (e.g. monotonicity via non-decreasing control points) in your ad surplus optimization work, and KAN-style layers.
I set out to keep it small, pure PyTorch, readable, and easy to maintain, with only modest speed ambitions. It is reasonably fast for such a library, and the performance comes mainly from leaning on PyTorch’s vectorization and efficient algorithms (Cox-De Boor for the splines, Clenshaw for the polynomials). The vectorized code can evaluate several curves at a batch of points in one call.
Gradients flow to both the inputs and the coefficients via custom autograd. It’s not the fastest implementation out there, but it stays simple and transparent.
Would love to that it works for you, and what would make it better. Have fun using it!
Imagine this guy interviewing you for your ML theory interview:
"Derive your loss from first principles."
"Stop, not the Bernoulli distribution, start by giving a definition of TRUTH "
I don't think there's anyone at NVIDIA having the rare combination of both engineering talent, business talent, connections, and vision. This is of course my personal belief.
People will continue "executing". But executing without innovating is meaningless - it will halt. A competitor that can innovate will win, but until that happens - progress is stuck.
@fromdevoid@soupcanarchist I want to enjoy great AI, innovation in scientific computing hardware, and my savings accounts not losing money. And I think the board achieved the opposite. This, in addition to the media ruining the lives of two NVIDIA women executives by spreading unconfirmed reports.
@soupcanarchist The "three other women" is a rumor. The only affair that was confirmed is between Huang and the executive from Megaspeed. You all are treating the other "alleged" affairs as if it's a fact, where we don't know it to be, and ruining the lives of these two women based on rumors!
Remember reading plenty of Cremers' work when working on image segmentation through first-order methods. Also attended some talks. A truly inspirational figure.
@KevHarron@soupcanarchist As we already established, the big nono is just a rumor that you're all spreading. These two NVIDIA executives have their lives ruined by the likes of you.
@KevHarron@soupcanarchist No he hasn't. That's why it's just a rumor. And you're all spreading an evil rumor about two executives @ NVIDIA and ruining their lives.
The only admission was the relationship with the executive from Megaspeed. The alleged Chinese spy.
@KevHarron@soupcanarchist Beyond a rumor in WSJ, is there an investigation that established that?
And why are you speaking about a rumor as if it's a fact? You're all ruining the lives of the two female executives. Imagine how they and their families feel now. What their kids are going through.
@AIQuanting How about the length of the correct prefix? Perhaps I can't guess the right prime factor, but a correct prefix gives me a narrow interval to search.
You are training a machine-learned model to predict the smallest prime factor of semi-primes.
Input: some representation of a semi-prime.
Output: the bits of its smallest prime factor.
What metrics do you measure on the validation / test sets that give a notion of progress?
🔥︎Delighted, thrilled, happy and please to share torchcurves - parametric curves in pure PyTorch. I believe it's mature enough to share here. Inspired by a two-year old paper from Yahoo.
code: https://t.co/MkW8DDKamf
docs: https://t.co/7jQlufbZch
paper that inspired it: https://t.co/FuFbqX8VSI
Use:
pip install torchcurves
or my personal preference:
uv add torchcurves
A small library of vectorized, differentiable B-spline and Legendre curves with learnable coefficients. Handy for continuous embeddings of numerical features in your recommender system, shape-constrained models (e.g. monotonicity via non-decreasing control points) in your ad surplus optimization work, and KAN-style layers.
I set out to keep it small, pure PyTorch, readable, and easy to maintain, with only modest speed ambitions. It is reasonably fast for such a library, and the performance comes mainly from leaning on PyTorch’s vectorization and efficient algorithms (Cox-De Boor for the splines, Clenshaw for the polynomials). The vectorized code can evaluate several curves at a batch of points in one call.
Gradients flow to both the inputs and the coefficients via custom autograd. It’s not the fastest implementation out there, but it stays simple and transparent.
Would love to that it works for you, and what would make it better. Have fun using it!
@fire622263921@witte_sergei Это в совке есть кварталы для ядерщиков. А Россия тот же совок. У нас в свободном Израиле каждый живёт там где ему хочется и нравиться. Только военнослужащие спят на базах. У вас просто нет представления свободной страны. Does not compute.
@giladmp@LazarFibich@naamalazimi זו הכללה גסה מאד. זה כבר תלוי ביחסים בין ההורה לילד וכמה ההורה נותן לו. וכמה הילד בכלל מוכן לקבל או שואף להיות בזכות עצמו.
זו הצעת חוק פופוליסטית מאד שמענישה את אלו שנתנו לילדיהם חינוך טוב כדי להגיע לאילו שנתנו לילדיהם חינוך רע.