홍석쓰 블로그

요약: Non-parametric statistics은 histogram 같이 데이터의 분포를 고정된 수의 parameter (ex. normal distribution model with mean, variance)로 표현하지 않는 통계방식을 의미한다. 이 포스트는 밑의 글을 번역해봤다. https://www.investopedia.com/terms/n/nonparametric-statistics.asp How Nonparametric Statistics Work Nonparametric statistics refer to a statistical method in which the data is not required to fit a normal distribution. Rankings should ..

https://deepgenerativemodels.github.io/notes/ Contents deepgenerativemodels.github.io 이런 수업을 학부 때 들을 수 있다는 게 부럽다. 스탠포드와 서울대의 수준 차이가 정말 까마득하구나. 입학할 때 학생 수준은 비슷하거나 더 좋을텐데, 졸업할 때는 하하... 지금 일단 intro, autoregressive까지 봤고, 원래 목적인 normalizing flow를 읽을건데, 여기 내용들을 전부 한 번 더 보면서 번역해 블로그에 쓰면 내 성장에 도움이 될 것 같다.

https://discuss.pytorch.org/t/loss-backward-failure-due-to-contiguous-issue/96416/19 Loss.backward() failure due to contiguous issue If that comes from here, if you just run the previous layer and this one, l10, does that trigger the same error? 🙂 discuss.pytorch.org feature를 layer마다 쌓아놨다가 concat하면 이런 에러가 생긴다. pixelnerrf에서는 이런에러가 없었는데.. 왜지?

경식이형 코드임. 1. cd rasterize 2. rm -d -r -f build 3. python setup.py build 4. python setup.py install 5. cd .. 6. python get_face_idx_map.py 이렇게하면 내가 첨부한 face_idx_map.jpg가 나와야함

2021.12.27 rasterize할 때 쓰려고 봤다가 기억 잘 안나는 거 보니 결국 안 쓴 것 같음. https://pytorch-scatter.readthedocs.io/en/1.3.0/functions/min.html Scatter Min — pytorch_scatter 1.3.0 documentation Parameters: src (Tensor) – The source tensor. index (LongTensor) – The indices of elements to scatter. dim (int, optional) – The axis along which to index. (default: -1) out (Tensor, optional) – The destination tensor. (de..

https://stackoverflow.com/questions/37987839/how-can-i-run-tensorboard-on-a-remote-server How can I run Tensorboard on a remote server? I'm new to Tensorflow and would greatly benefit from some visualizations of what I'm doing. I understand that Tensorboard is a useful visualization tool, but how do I run it on my remote Ubuntu mac... stackoverflow.com

Udate 2021.08.24. 내가말한게 이거였음 https://github.com/pytorch/pytorch/issues/5059#issuecomment-404232359 np.random generates the same random numbers for each data batch · Issue #5059 · pytorch/pytorch For example, the following script from __future__ import print_function import numpy as np import random # np.random.seed(0) class Transform(object): def __init__(self): pass def __call__(self, ite... gi..

Update: 02/18/2024 https://huggingface.co/docs/transformers/v4.20.1/en/perf_train_gpu_many#:~:text=DP%20copies%20data%20within%20the,not%20the%20case%20with%20DP. ————/ DDP를 써라! 핵심은 DP는 single-process multi-thread이고, DDP는 multi-process라는 것. python GIL overhead를 피하려면 multi-thread가 아닌 multi-process를 쓰는게 효율적임.https://pytorch.org/docs/master/notes/cuda.html#use-nn-parallel-distributeddataparallel-in..

Update 2021.12.27 torch로 train할 때 segmentation fault (core dump) 버그가 뜨는데, osmesa import할 때만 떴음. 원인은 아직 불명. Update 2021.08.16 the problem was osmesa. egl has no problem os.environ['PYOPENGL_PLATFORM'] = 'egl' but normally osmesa has no bugs... what should I do python debugger sometimes gives segmentation fault. Most of cases seem to be related to dependencies between python libraries. In my case,..

전에 코드는 잘이해안갔는데, 이건 바로 이해감. https://github.com/zju3dv/neuralbody/blob/master/lib/utils/if_nerf/if_nerf_data_utils.py def get_near_far(bounds, ray_o, ray_d): """calculate intersections with 3d bounding box""" norm_d = np.linalg.norm(ray_d, axis=-1, keepdims=True) viewdir = ray_d / norm_d viewdir[(viewdir -1e-10)] = 1e-5 viewdir[(viewdir > -1e-5) & (viewdir < 1e-10)] = -1e-5 tm..