Hansol Kang, profile picture
Uploaded byHansol Kang
PDF, PPTX1,421 views

PyTorch 튜토리얼 (Touch to PyTorch)

PyTorch is an open-source machine learning library for Python. It is primarily developed by Facebook's AI research group. The document discusses setting up PyTorch, including installing necessary packages and configuring development environments. It also provides examples of core PyTorch concepts like tensors, common datasets, and constructing basic neural networks.

Technology◦
Related topics:
Deep Learning•

Embed presentation

Download as PDF, PPTX
Touch to PyTorch ISL Lab Seminar Hansol Kang : From basic to vanilla GAN
Contents October 9, 2018 Setup Install Development Tools Example What is PyTorch? PyTorch Deep Learning Framework Tensor Datasets Neural Nets Learning Applications
What is PyTorch? October 9, 2018 2 an open source machine learning library for Python. It is primarily developed by Facebook's artificial-intelligence research group. • PyTorch Vs.
What is PyTorch? October 9, 2018 3 Written in : Python Interface : PythonNov. 2010 Written in : C++ Interface : Python, MATLAB, C++Dec. 2013 Written in : C, Lua Interface : C, LuaJul. 2014 Written in : Python Interface : Python, RMar. 2015 Written in : C++, Python, CUDA Interface : Python, C/C++, Java, Go, R, JuliaNov. 2015 Written in : Interface : Python, C++Apr. 2017 Written in : Python, C, CUDA Interface : PythonOct. 2016 ** DL4J(Java) Chainer(Python) MXNet(C++, Python, Julia, MATLAB, JavaScript, Go, R, Scala, Perl) CNTK(Python, C++), TF Learn(Python) TF-Slim(Python) Etc. Recommend to choose these framework • Deep Learning Framework “What is this? Gum? It's GAN.”, pp. 21-22 Cf.
Setup October 9, 2018 4 • Install https://www.anaconda.com/download/
Setup October 9, 2018 5 • Install https://pytorch.org/ C++ : Preview, Linux only
Setup October 9, 2018 6 • Install conda create -n PyTorch python=3.6 activate PyTorch conda install pytorch cuda90 -c pytorch pip install torchvision o conda create는 환경 생성하는 명령어. PyTorch 뿐만 아니라 Tensorflow 같은 다른 딥러닝 프레임워크를 사용한다거나 하위 파이썬 버전도 사용해야하는 경우 환경마다 설정해주면, 디펜던시가 꼬이지 않음. o -n 환경명, python=파이썬버전 입력. 환경설정 리스트는 conda env list를 입력하면 확인 가능. o activate는 해당 환경을 활성화 시키는 명령어. 반대로 환경을 빠져나오는 명령어는 deactivate. o PyTorch를 설치하는 명령어는 conda install pytorch cuda90 -c pytorch.. o torchvision은 딥러닝 학습에 많이 사용되는 데이터셋, 네트워크 구조, 이미지 변환과 같은 기능을 제공하므로 설치하는 것을 권장.
Setup October 9, 2018 7 • Development Tools : Cell based execution : Intellisense : Intellisense : Management of python env. : GitHub : AI tool package : Startup file : Intellisense : Cell based execution : Management of python env. : Extension program : Insane extension program : Intellisense : Environment setting
Setup October 9, 2018 8 • Development Tools Personal opinion : >
Setup October 9, 2018 9 • Development Tools – Visual Studio File-New-Project
Setup October 9, 2018 10 • Development Tools – Visual Studio File-New-Project Python 도구 설치
Setup October 9, 2018 11 • Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application
Setup October 9, 2018 12 • Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가
Setup October 9, 2018 13 • Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가 코드 작성
Setup October 9, 2018 14 • Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가 코드 작성
Setup October 9, 2018 15 • Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가 코드 작성 실행 : 시작 파일로 설정-> Ctrl+F5
Setup October 9, 2018 16 • Development Tools – PyCharm
Setup October 9, 2018 17 • Development Tools – PyCharm
Setup October 9, 2018 18 • Development Tools – PyCharm
Setup October 9, 2018 19 • Development Tools – PyCharm
Setup October 9, 2018 20 • Development Tools – PyCharm
Setup October 9, 2018 21 • Development Tools – PyCharm File-New Project
Setup October 9, 2018 22 • Development Tools – PyCharm File-New Project 코드 작성
Setup October 9, 2018 23 • Development Tools – PyCharm File-New Project 코드 작성 실행 : Ctrl+F5 “Too easy, but important things”, pp. 42-43 Cf. TODO
Example October 9, 2018 24 • Tensor array Tensor reshape view(reshape) linspace linspace ones, zeros ones, zeros random.randn, random.rand randn, rand from_numpy numpy to, cuda
Example October 9, 2018 25 • Tensor import torch as tc #Tensor a = tc.Tensor([2.5, 4]) b = tc.Tensor([[1, 2.5], [2.5, 6]]) #reshape c = b.reshape(1, 4) d = b.reshape(1, -1) Tensor a : tensor([ 2.5000, 4.0000]) Size of a : torch.Size([2]) Tensor b : tensor([[ 1.0000, 2.5000], [ 2.5000, 6.0000]]) Size of b : torch.Size([2, 2]) Tensor c : tensor([[ 1.0000, 2.5000, 2.5000, 6.0000]]) Size of c : torch.Size([1, 4]) Tensor d : tensor([[ 1.0000, 2.5000, 2.5000, 6.0000]]) Size of d : torch.Size([1, 4])
Example October 9, 2018 26 • Tensor Start End 0 #ones, zeros real = tc.ones(2, 2) fake = tc.zeros(2, 2) 1 #linspace x = tc.linspace(-1, 1, 9) Tensor x : tensor([-1.0000, -0.7500, -0.5000, -0.2500, 0.0000, 0.2500, 0.5000, 0.7500, 1.0000]) Size of x : torch.Size([9]) Tensor real : tensor([[ 1., 1.], [ 1., 1.]]) Size of real : torch.Size([2, 2]) Tensor fake : tensor([[ 0., 0.], [ 0., 0.]]) Size of fake : torch.Size([2, 2])
Example October 9, 2018 27 • Tensor 0 1 0 1 #randn, rand z1 = tc.randn(2, 3) z2 = tc.rand(2, 3) Tensor z1 : tensor([[-0.0750, -1.0555, -0.0706], [-0.1946, -0.8593, -0.2238]]) Size of z1 : torch.Size([2, 3]) Tensor z2 : tensor([[ 0.8204, 0.3505, 0.1034], [ 0.5318, 0.9543, 0.8781]]) Size of z2 : torch.Size([2, 3]) #to, cuda z_cuda = z1.cuda() device = tc.device('cuda' if tc.cuda.is_available() else 'cpu') z_device = z1.to(device) Tensor z_cuda : tensor([[-0.0750, -1.0555, -0.0706], [-0.1946, -0.8593, -0.2238]], device='cuda:0') Tensor z_device : tensor([[-0.0750, -1.0555, -0.0706], [-0.1946, -0.8593, -0.2238]], device='cuda:0')
Example October 9, 2018 28 • Tensor #from_numpy, numpy import numpy as np a = np.array([3.5, 4]) b = tc.from_numpy(a) c = b.numpy() array a : [3.5 4. ] Tensor b : tensor([ 3.5000, 4.0000], dtype=torch.float64) array c : [3.5 4. ]
Example October 9, 2018 29 • Datasets The following datasets are available: MNIST Fashion-MNIST EMNIST COCO LSUN ImageFolder DatasetFolder Imagenet-12 CIFAR STL10 SVHN PhotoTour : 바로 사용 가능. : 추가적인 과정 필요. : Custom dataset에 활용. MNIST Fashion-MNIST CIFAR LSUN CelebA K-pop(Custom)
Example October 9, 2018 30 • Datasets – MNIST trans  Compose()is used when there are multiple transform options. Here, ToTensor() and Normalize(mean, std) are used.  ToTensor () changes the PIL Image to a tensor. torchvision dataset The default type is PIL Image.  Normalize (mean, std) transforms the range of the image. Here, the value of [0, 1] is adjusted to [-1, 1]. ((value-mean) / std) trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) import torch as tc import torchvision as tv import torchvision.transforms as transforms
Example October 9, 2018 31 • Datasets – MNIST dataset  root : This is the path to store (MNIST data). Folders are automatically created with the specified name.  train : Set the data to be used for the train.  transform : Transform the data according to the transform option set previously.  download : Download (MINST data). (If you downloaded it once, it will not do it again.) trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) import torch as tc import torchvision as tv import torchvision.transforms as transforms
Example October 9, 2018 32 • Datasets – MNIST dataloader  dataset : Set the dataset to load.  batch_size : Set the batch size.  shuffle : Shuffle the data and load it. trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) import torch as tc import torchvision as tv import torchvision.transforms as transforms
Example October 9, 2018 33 • Datasets – LSUN https://github.com/fyu/lsun
Example October 9, 2018 34 • Datasets – LSUN Run download.py def list_categories(tag): url = 'http://lsun.cs.princeton.edu/htbin/list.cgi?tag=' + tag f = urlopen(url) return json.loads(f.read()) If you are using Python 3.0 or later, modify the code from urllib2.urlopen (url) to urlopen (url). download.py -c bedroom bedroom 0 bridge 1 church_outdoor 2 classroom 3 conference_room 4 dining_room 5 kitchen 6 living_room 7 restaurant 8 tower 9
Example October 9, 2018 35 • Datasets – LSUN trans = transforms.Compose([transforms.Resize((64,64)) ,transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.LSUN('.', classes=['bedroom_train'], transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) trans  Resize() is used to resize the image. datasets  root : Root directory for the database files.  classes : One of {‘train’, ‘val’, ‘test’} or a list of categories to load. e,g. [‘bedroom_train’, ‘church_train’].
Example October 9, 2018 36 • Datasets – ImageFolder(CelebA) http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
Example October 9, 2018 37 • Datasets – ImageFolder(CelebA) http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html 압축을 풀면 jpg 형태로 파일들이 저장된 것을 확인할 수 있음.
Example October 9, 2018 38 • Datasets – ImageFolder(CelebA) trans = transforms.Compose([transforms.Resize((64,64)) ,transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.ImageFolder('./img_align_celeba', trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) datasets  root : Root directory for the database files.  transform : Transform the data according to the transform option set previously.
H Example October 9, 2018 39 • Neural Nets Sequence baseClass base model = tc.nn.Sequential( tc.nn.Linear(D_in, H), tc.nn.ReLU(), tc.nn.Linear(H, D_out), tc.nn.Sigmoid() ) class Model(tc.nn.Module): def __init__(self): super(Model, self).__init__() self.linear1 = tc.nn.Linear(D_in, H), self.linear2 = tc.nn.Linear(H, D_out) def forward(self, input): x = tc.nn.functional.relu(self.linear1(input)) x = tc.nn.functional.sigmoid(self.linear2(x)) return x D_in D_out Data manipulation
Example October 9, 2018 40 • Neural Nets Class base class Model(tc.nn.Module): def __init__(self): super(Model, self).__init__() self.linear1 = tc.nn.Linear(D_in, H), self.linear2 = tc.nn.Linear(H, D_out) def forward(self, input): x = tc.nn.functional.relu(self.linear1(input)) x = tc.nn.functional.sigmoid(self.linear2(x)) return x tc.nn.Module functions Model functions 생성자 부모 클래스 초기화 Propagation Cf. Diamond problem A B C D Class D __init__() Class B __init__() Class A __init__() Class C __init__() Class A __init__()
Example October 9, 2018 41 • Learning Back Propagation (Learning) Loss Optimizer loss_func = tc.nn.MSELoss() opt = tc.optim.Adam(model.parameters(), lr=0.01 ) for ep in range(epoch_sz): for step, (images, labels) in enumerate(dataloader): opt.zero_grad() images = images.to(device) labels = labels.to(device) output = model(images) loss = loss_func(output) loss.backward() opt.step()
Example October 9, 2018 42 • Learning loss_func = tc.nn.MSELoss() opt = tc.optim.Adam(model.parameters(), lr=0.01 ) for ep in range(epoch_sz): for step, (images, labels) in enumerate(dataloader): opt.zero_grad() images = images.to(device) labels = labels.to(device) output = model(images) loss = loss_func(output) loss.backward() opt.step() Loss 선언 Optimizer 선언 (업데이트 하려는 parameter) 정해진 epoch 만큼 수행 앞서 설정한 dataloader에 따라 image(data)와 label(_) 불러옴 Gradient 초기화. Loss 구함 Back propagation(Gradient 구함) 정해진 optimizer에 따라 parameter 업데이트 trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) Cf.
Example October 9, 2018 43 • Learning • L1Loss • MSELoss • CrossEntropyLoss • NLLLoss • PoissonNLLLoss • KLDivLoss • BCELoss • BCEWithLogitsLoss • MarginRankingLoss • HingeEmbeddingLoss • SmoothL1Loss • SoftMarginLoss • MultiLabelSoftMarginLoss • CosineEmbeddingLoss • MultiMarginLoss • TripleMarginLoss • Adadelta • Adagrad • Adam • SparseAdam • Adamax • ASGD • LBFGS • RMSprop • Rprop • SGD Loss Optimizer * Bold체는 주로 사용하는 함수들.
Example • Applications –Vanilla GAN 2018-10-09 44 https://github.com/messy-snail/GAN_PyTorch “What is this? Gum? It's GAN.” Cf.
Example • Applications –Vanilla GAN 2018-10-09 45 Sequence baseClass base Neural Nets
Example • Applications –Vanilla GAN 2018-10-09 46
Example • Applications –Vanilla GAN 2018-10-09 47 Images(batch=100) … 28 28 100 Reshape(28*28*1->784) 7 8 4 7 8 4… 7 8 4 100 1 0 0 Noise(z) Label 1 0 Real Fake
Example • Applications –Vanilla GAN 2018-10-09 48 1 0 0 Noise(z) Label 1 0 Real Fake G (z) Generate D(Fakes) D(Images) Discriminate Loss Real Loss Fake Loss Reshape(28*28*1->784) 7 8 4 7 8 4… 7 8 4 100
Example • Applications –Vanilla GAN 2018-10-09 49 D(Fakes) D(Images) Discriminate Loss Real Loss Fake Total Loss Loss Update
Example • Applications –Vanilla GAN 2018-10-09 50 1 0 0 Noise(z) Label 1 Real G (z) Generate D(Fakes) Discriminate Loss g Loss Update
Future Work October 9, 2018 51 Paper Review Vanilla GAN DCGAN InfoGAN Unrolled GAN Wasserstein GAN LS GAN BEGAN Pix2Pix Cycle GAN Proposed Model SpyGAN Tools Document Programming PyTorch Python executable & UI Mathematical Study Linear algebra Probability and statistics Information theory Others Level Processor Ice Propagation Maybe next seminar?
52 &

More Related Content

PDF
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
byHansol Kang
 
PDF
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
byHansol Kang
 
PDF
Deep Convolutional GANs - meaning of latent space
byHansol Kang
 
PDF
Deep Learning with PyTorch
byMayur Bhangale
 
PDF
PyTorch for Deep Learning Practitioners
byBayu Aldi Yansyah
 
PPTX
Introduction to PyTorch
byJun Young Park
 
PDF
05 Analysis of Algorithms: Heap and Quick Sort - Corrected
byAndres Mendez-Vazquez
 
PDF
Speaker Diarization
byHONGJOO LEE
 
쉽게 설명하는 GAN (What is this? Gum? It's GAN.)
byHansol Kang
 
딥러닝 중급 - AlexNet과 VggNet (Basic of DCNN : AlexNet and VggNet)
byHansol Kang
 
Deep Convolutional GANs - meaning of latent space
byHansol Kang
 
Deep Learning with PyTorch
byMayur Bhangale
 
PyTorch for Deep Learning Practitioners
byBayu Aldi Yansyah
 
Introduction to PyTorch
byJun Young Park
 
05 Analysis of Algorithms: Heap and Quick Sort - Corrected
byAndres Mendez-Vazquez
 
Speaker Diarization
byHONGJOO LEE
 

What's hot

PDF
Pytorch for tf_developers
byAbdul Muneer
 
PDF
Python NumPy Tutorial | NumPy Array | Edureka
byEdureka!
 
PPTX
Aaex3 group2
byShiang-Yun Yang
 
PPTX
Transfer learning, active learning using tensorflow object detection api
by설기 김
 
PDF
"PyTorch Deep Learning Framework: Status and Directions," a Presentation from...
byEdge AI and Vision Alliance
 
PPTX
Scaling Python to CPUs and GPUs
byTravis Oliphant
 
PDF
Large scale logistic regression and linear support vector machines using spark
byMila, Université de Montréal
 
PDF
Real-Time Big Data Stream Analytics
byAlbert Bifet
 
PDF
PyCon Estonia 2019
byTravis Oliphant
 
PDF
Real Time Big Data Management
byAlbert Bifet
 
PDF
Automatic Features Generation And Model Training On Spark: A Bayesian Approach
bySpark Summit
 
PDF
Sea Amsterdam 2014 November 19
byGoDataDriven
 
PDF
Keynote at Converge 2019
byTravis Oliphant
 
PPTX
Scaling out logistic regression with Spark
byBarak Gitsis
 
PDF
2014-06-20 Multinomial Logistic Regression with Apache Spark
byDB Tsai
 
PDF
Working with the Scalding Type-Safe API
byCriteolabs
 
PPT
Pythonic Math
byKirby Urner
 
PDF
The Joy of SciPy
bykammeyer
 
PPTX
CppConcurrencyInAction - Chapter07
byDooSeon Choi
 
PDF
DPF 2017: GPUs in LHCb for Analysis
byHenry Schreiner
 
Pytorch for tf_developers
byAbdul Muneer
 
Python NumPy Tutorial | NumPy Array | Edureka
byEdureka!
 
Aaex3 group2
byShiang-Yun Yang
 
Transfer learning, active learning using tensorflow object detection api
by설기 김
 
"PyTorch Deep Learning Framework: Status and Directions," a Presentation from...
byEdge AI and Vision Alliance
 
Scaling Python to CPUs and GPUs
byTravis Oliphant
 
Large scale logistic regression and linear support vector machines using spark
byMila, Université de Montréal
 
Real-Time Big Data Stream Analytics
byAlbert Bifet
 
PyCon Estonia 2019
byTravis Oliphant
 
Real Time Big Data Management
byAlbert Bifet
 
Automatic Features Generation And Model Training On Spark: A Bayesian Approach
bySpark Summit
 
Sea Amsterdam 2014 November 19
byGoDataDriven
 
Keynote at Converge 2019
byTravis Oliphant
 
Scaling out logistic regression with Spark
byBarak Gitsis
 
2014-06-20 Multinomial Logistic Regression with Apache Spark
byDB Tsai
 
Working with the Scalding Type-Safe API
byCriteolabs
 
Pythonic Math
byKirby Urner
 
The Joy of SciPy
bykammeyer
 
CppConcurrencyInAction - Chapter07
byDooSeon Choi
 
DPF 2017: GPUs in LHCb for Analysis
byHenry Schreiner
 

Similar to PyTorch 튜토리얼 (Touch to PyTorch)

PDF
How to install AI open source tool and environment
byTae wook kang
 
PDF
PyTorch crash course
byNader Karimi
 
PDF
Icpp power ai-workshop 2018
byGanesan Narayanasamy
 
PDF
pytdddddddddddddddddddddddddddddddddorch.pdf
bydrjigarsoni28
 
PDF
Machine learning with py torch
byRiza Fahmi
 
PDF
OpenPOWER Workshop in Silicon Valley
byGanesan Narayanasamy
 
PPTX
Tutorial: Image Generation and Image-to-Image Translation using GAN
byWuhyun Rico Shin
 
PDF
Deep-Learning-with-PydddddddddddddTorch.pdf
bydrjigarsoni28
 
PPTX
Transfer Leaning Using Pytorch synopsis Minor project pptx
byAnkit Gupta
 
PPTX
Deep learning study 3
bySan Kim
 
PPTX
Soumith Chintala - Increasing the Impact of AI Through Better Software
byMLconf
 
PDF
오픈소스로 시작하는 인공지능 실습
byMario Cho
 
PDF
Koss 1605 machine_learning_mariocho_t10
byMario Cho
 
PPTX
Demystifying-AI-Frameworks-TensorFlow-PyTorch-JAX-and-More (1).pptx
byAnant Garg
 
PDF
From NumPy to PyTorch
byMike Ruberry
 
PDF
Simple, fast, and scalable torch7 tutorial
byJin-Hwa Kim
 
PDF
PyTorch Introduction
byYash Kawdiya
 
PPTX
pytorch_tutorial_follow_this_to_start.pptx
bygyungmindenniskim
 
PDF
1-bit semantic segmentation
byJeonghoonKim30
 
PDF
Pytorch A Detailed Overview Agladze Mikhail
byilzobrzan47
 
How to install AI open source tool and environment
byTae wook kang
 
PyTorch crash course
byNader Karimi
 
Icpp power ai-workshop 2018
byGanesan Narayanasamy
 
pytdddddddddddddddddddddddddddddddddorch.pdf
bydrjigarsoni28
 
Machine learning with py torch
byRiza Fahmi
 
OpenPOWER Workshop in Silicon Valley
byGanesan Narayanasamy
 
Tutorial: Image Generation and Image-to-Image Translation using GAN
byWuhyun Rico Shin
 
Deep-Learning-with-PydddddddddddddTorch.pdf
bydrjigarsoni28
 
Transfer Leaning Using Pytorch synopsis Minor project pptx
byAnkit Gupta
 
Deep learning study 3
bySan Kim
 
Soumith Chintala - Increasing the Impact of AI Through Better Software
byMLconf
 
오픈소스로 시작하는 인공지능 실습
byMario Cho
 
Koss 1605 machine_learning_mariocho_t10
byMario Cho
 
Demystifying-AI-Frameworks-TensorFlow-PyTorch-JAX-and-More (1).pptx
byAnant Garg
 
From NumPy to PyTorch
byMike Ruberry
 
Simple, fast, and scalable torch7 tutorial
byJin-Hwa Kim
 
PyTorch Introduction
byYash Kawdiya
 
pytorch_tutorial_follow_this_to_start.pptx
bygyungmindenniskim
 
1-bit semantic segmentation
byJeonghoonKim30
 
Pytorch A Detailed Overview Agladze Mikhail
byilzobrzan47
 

More from Hansol Kang

PDF
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
byHansol Kang
 
PDF
Support Vector Machine - 기본 이해와 OpenCV 실습.pdf
byHansol Kang
 
PDF
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
byHansol Kang
 
PPTX
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
byHansol Kang
 
PDF
알아두면 쓸모있는 깃허브 2
byHansol Kang
 
PDF
알아두면 쓸모있는 깃허브 1
byHansol Kang
 
PDF
FPN 리뷰
byHansol Kang
 
PDF
R-FCN 리뷰
byHansol Kang
 
PDF
basic of deep learning
byHansol Kang
 
PDF
파이썬 제대로 활용하기
byHansol Kang
 
PPTX
모던 C++ 정리
byHansol Kang
 
PDF
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
byHansol Kang
 
PDF
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
byHansol Kang
 
PDF
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
byHansol Kang
 
PDF
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
byHansol Kang
 
PDF
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
byHansol Kang
 
PDF
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
byHansol Kang
 
PDF
QT 프로그래밍 기초(basic of QT programming tutorial)
byHansol Kang
 
PDF
Continuously Adaptive Mean Shift(CAMSHIFT)
byHansol Kang
 
PDF
Mobile Robot PD and DOB control
byHansol Kang
 
이 세계로의 전송_파이썬과 함께하는 궤도모험.pdf
byHansol Kang
 
Support Vector Machine - 기본 이해와 OpenCV 실습.pdf
byHansol Kang
 
ROS 시작하기(Getting Started with ROS:: Your First Steps in Robot Programming )
byHansol Kang
 
관측 임무스케줄링 (Selecting and scheduling observations of agile satellites)
byHansol Kang
 
알아두면 쓸모있는 깃허브 2
byHansol Kang
 
알아두면 쓸모있는 깃허브 1
byHansol Kang
 
FPN 리뷰
byHansol Kang
 
R-FCN 리뷰
byHansol Kang
 
basic of deep learning
byHansol Kang
 
파이썬 제대로 활용하기
byHansol Kang
 
모던 C++ 정리
byHansol Kang
 
Photo-realistic Single Image Super-resolution using a Generative Adversarial ...
byHansol Kang
 
LSGAN - SIMPle(Simple Idea Meaningful Performance Level up)
byHansol Kang
 
InfoGAN : Interpretable Representation Learning by Information Maximizing Gen...
byHansol Kang
 
문서와 개발에 필요한 간단한 팁들(Too easy, but important things - document, development)
byHansol Kang
 
신뢰 전파 기법을 이용한 스테레오 정합(Stereo matching using belief propagation algorithm)
byHansol Kang
 
HSV 컬러 공간에서의 레티넥스와 채도 보정을 이용한 화질 개선 기법
byHansol Kang
 
QT 프로그래밍 기초(basic of QT programming tutorial)
byHansol Kang
 
Continuously Adaptive Mean Shift(CAMSHIFT)
byHansol Kang
 
Mobile Robot PD and DOB control
byHansol Kang
 

Recently uploaded

PDF
How Mobile Apps Are Shaping the Future of Digital Innovation
byWilliam Taylor
 
PDF
eResource Scheduler Enterprise Resource Management and Scheduling Software.pdf
byeResource Scheduler
 
PDF
Ransomware_Resilience_Strategic_Playbook.pdf
byAfonso Henrique Rodrigues Alves
 
PDF
Vibe Coding vs. Spec-Driven Development [Free Meetup]
byHaim Michael
 
PPTX
Basics of Identity Access Management In mordern Infrastructure
byPrinceXavier18
 
PDF
Guided Substation Engineering with CoMPAS: Simplifying IEC 61850
byDanBrown980551
 
DOCX
Introduction to the World of Computers (Hardware & Software)
byALIABBAS ABASOV
 
PPTX
RISE with SAP for Automotive - S4 HANA Value.pptx
byAmira Jemi
 
PDF
AI Technology important knowledge ......
byutkarshj2007
 
PDF
Generative AI in 2026: Hype, Bubble, Winter or The Real Deal?
byDr. Tathagat Varma
 
PDF
TrustArc Webinar - Assessing AI Risk with Confidence
byTrustArc
 
PDF
Innovative AI Solutions for Business Growth.pdf
byVox Works
 
PDF
AI and Computer Architecture: 200 Years Together
byDmitry Zinoviev
 
PDF
DevSecOps Learning Path - TryHackMe Certificate
byVICTOR MAESTRE RAMIREZ
 
PDF
DIGITAL FORENSICS - Notes for Everything.pdf
bypankajkumavatbeit
 
PDF
Decoding the DNA: The Digital Networks Act, the Open Internet, and IP interco...
byCSUC - Consorci de Serveis Universitaris de Catalunya
 
PPTX
Combining Induction and Transduction for Abstract Reasoning.pptx
byallen578468
 
PDF
Taming the Chaos: How to Turn Unstructured Data into Decisions
bySafe Software
 
PDF
Making Sense of Raster: From Bit Depth to Better Workflows
bySafe Software
 
PDF
Top 7 Manufacturing Software for Small Businesses Boosting Growth in 2025
byEnvertis Software Solutions
 
How Mobile Apps Are Shaping the Future of Digital Innovation
byWilliam Taylor
 
eResource Scheduler Enterprise Resource Management and Scheduling Software.pdf
byeResource Scheduler
 
Ransomware_Resilience_Strategic_Playbook.pdf
byAfonso Henrique Rodrigues Alves
 
Vibe Coding vs. Spec-Driven Development [Free Meetup]
byHaim Michael
 
Basics of Identity Access Management In mordern Infrastructure
byPrinceXavier18
 
Guided Substation Engineering with CoMPAS: Simplifying IEC 61850
byDanBrown980551
 
Introduction to the World of Computers (Hardware & Software)
byALIABBAS ABASOV
 
RISE with SAP for Automotive - S4 HANA Value.pptx
byAmira Jemi
 
AI Technology important knowledge ......
byutkarshj2007
 
Generative AI in 2026: Hype, Bubble, Winter or The Real Deal?
byDr. Tathagat Varma
 
TrustArc Webinar - Assessing AI Risk with Confidence
byTrustArc
 
Innovative AI Solutions for Business Growth.pdf
byVox Works
 
AI and Computer Architecture: 200 Years Together
byDmitry Zinoviev
 
DevSecOps Learning Path - TryHackMe Certificate
byVICTOR MAESTRE RAMIREZ
 
DIGITAL FORENSICS - Notes for Everything.pdf
bypankajkumavatbeit
 
Decoding the DNA: The Digital Networks Act, the Open Internet, and IP interco...
byCSUC - Consorci de Serveis Universitaris de Catalunya
 
Combining Induction and Transduction for Abstract Reasoning.pptx
byallen578468
 
Taming the Chaos: How to Turn Unstructured Data into Decisions
bySafe Software
 
Making Sense of Raster: From Bit Depth to Better Workflows
bySafe Software
 
Top 7 Manufacturing Software for Small Businesses Boosting Growth in 2025
byEnvertis Software Solutions
 

PyTorch 튜토리얼 (Touch to PyTorch)

  • 1.
    Touch to PyTorch ISLLab Seminar Hansol Kang : From basic to vanilla GAN
  • 2.
    Contents October 9, 2018 Setup Install DevelopmentTools Example What is PyTorch? PyTorch Deep Learning Framework Tensor Datasets Neural Nets Learning Applications
  • 3.
    What is PyTorch? October9, 2018 2 an open source machine learning library for Python. It is primarily developed by Facebook's artificial-intelligence research group. • PyTorch Vs.
  • 4.
    What is PyTorch? October9, 2018 3 Written in : Python Interface : PythonNov. 2010 Written in : C++ Interface : Python, MATLAB, C++Dec. 2013 Written in : C, Lua Interface : C, LuaJul. 2014 Written in : Python Interface : Python, RMar. 2015 Written in : C++, Python, CUDA Interface : Python, C/C++, Java, Go, R, JuliaNov. 2015 Written in : Interface : Python, C++Apr. 2017 Written in : Python, C, CUDA Interface : PythonOct. 2016 ** DL4J(Java) Chainer(Python) MXNet(C++, Python, Julia, MATLAB, JavaScript, Go, R, Scala, Perl) CNTK(Python, C++), TF Learn(Python) TF-Slim(Python) Etc. Recommend to choose these framework • Deep Learning Framework “What is this? Gum? It's GAN.”, pp. 21-22 Cf.
  • 5.
    Setup October 9, 2018 4 •Install https://www.anaconda.com/download/
  • 6.
    Setup October 9, 2018 5 •Install https://pytorch.org/ C++ : Preview, Linux only
  • 7.
    Setup October 9, 2018 6 •Install conda create -n PyTorch python=3.6 activate PyTorch conda install pytorch cuda90 -c pytorch pip install torchvision o conda create는 환경 생성하는 명령어. PyTorch 뿐만 아니라 Tensorflow 같은 다른 딥러닝 프레임워크를 사용한다거나 하위 파이썬 버전도 사용해야하는 경우 환경마다 설정해주면, 디펜던시가 꼬이지 않음. o -n 환경명, python=파이썬버전 입력. 환경설정 리스트는 conda env list를 입력하면 확인 가능. o activate는 해당 환경을 활성화 시키는 명령어. 반대로 환경을 빠져나오는 명령어는 deactivate. o PyTorch를 설치하는 명령어는 conda install pytorch cuda90 -c pytorch.. o torchvision은 딥러닝 학습에 많이 사용되는 데이터셋, 네트워크 구조, 이미지 변환과 같은 기능을 제공하므로 설치하는 것을 권장.
  • 8.
    Setup October 9, 2018 7 •Development Tools : Cell based execution : Intellisense : Intellisense : Management of python env. : GitHub : AI tool package : Startup file : Intellisense : Cell based execution : Management of python env. : Extension program : Insane extension program : Intellisense : Environment setting
  • 9.
    Setup October 9, 2018 8 •Development Tools Personal opinion : >
  • 10.
    Setup October 9, 2018 9 •Development Tools – Visual Studio File-New-Project
  • 11.
    Setup October 9, 2018 10 •Development Tools – Visual Studio File-New-Project Python 도구 설치
  • 12.
    Setup October 9, 2018 11 •Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application
  • 13.
    Setup October 9, 2018 12 •Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가
  • 14.
    Setup October 9, 2018 13 •Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가 코드 작성
  • 15.
    Setup October 9, 2018 14 •Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가 코드 작성
  • 16.
    Setup October 9, 2018 15 •Development Tools – Visual Studio File-New-Project Python 도구 설치 Python Application Python 환경 추가 코드 작성 실행 : 시작 파일로 설정-> Ctrl+F5
  • 17.
    Setup October 9, 2018 16 •Development Tools – PyCharm
  • 18.
    Setup October 9, 2018 17 •Development Tools – PyCharm
  • 19.
    Setup October 9, 2018 18 •Development Tools – PyCharm
  • 20.
    Setup October 9, 2018 19 •Development Tools – PyCharm
  • 21.
    Setup October 9, 2018 20 •Development Tools – PyCharm
  • 22.
    Setup October 9, 2018 21 •Development Tools – PyCharm File-New Project
  • 23.
    Setup October 9, 2018 22 •Development Tools – PyCharm File-New Project 코드 작성
  • 24.
    Setup October 9, 2018 23 •Development Tools – PyCharm File-New Project 코드 작성 실행 : Ctrl+F5 “Too easy, but important things”, pp. 42-43 Cf. TODO
  • 25.
    Example October 9, 2018 24 •Tensor array Tensor reshape view(reshape) linspace linspace ones, zeros ones, zeros random.randn, random.rand randn, rand from_numpy numpy to, cuda
  • 26.
    Example October 9, 2018 25 •Tensor import torch as tc #Tensor a = tc.Tensor([2.5, 4]) b = tc.Tensor([[1, 2.5], [2.5, 6]]) #reshape c = b.reshape(1, 4) d = b.reshape(1, -1) Tensor a : tensor([ 2.5000, 4.0000]) Size of a : torch.Size([2]) Tensor b : tensor([[ 1.0000, 2.5000], [ 2.5000, 6.0000]]) Size of b : torch.Size([2, 2]) Tensor c : tensor([[ 1.0000, 2.5000, 2.5000, 6.0000]]) Size of c : torch.Size([1, 4]) Tensor d : tensor([[ 1.0000, 2.5000, 2.5000, 6.0000]]) Size of d : torch.Size([1, 4])
  • 27.
    Example October 9, 2018 26 •Tensor Start End 0 #ones, zeros real = tc.ones(2, 2) fake = tc.zeros(2, 2) 1 #linspace x = tc.linspace(-1, 1, 9) Tensor x : tensor([-1.0000, -0.7500, -0.5000, -0.2500, 0.0000, 0.2500, 0.5000, 0.7500, 1.0000]) Size of x : torch.Size([9]) Tensor real : tensor([[ 1., 1.], [ 1., 1.]]) Size of real : torch.Size([2, 2]) Tensor fake : tensor([[ 0., 0.], [ 0., 0.]]) Size of fake : torch.Size([2, 2])
  • 28.
    Example October 9, 2018 27 •Tensor 0 1 0 1 #randn, rand z1 = tc.randn(2, 3) z2 = tc.rand(2, 3) Tensor z1 : tensor([[-0.0750, -1.0555, -0.0706], [-0.1946, -0.8593, -0.2238]]) Size of z1 : torch.Size([2, 3]) Tensor z2 : tensor([[ 0.8204, 0.3505, 0.1034], [ 0.5318, 0.9543, 0.8781]]) Size of z2 : torch.Size([2, 3]) #to, cuda z_cuda = z1.cuda() device = tc.device('cuda' if tc.cuda.is_available() else 'cpu') z_device = z1.to(device) Tensor z_cuda : tensor([[-0.0750, -1.0555, -0.0706], [-0.1946, -0.8593, -0.2238]], device='cuda:0') Tensor z_device : tensor([[-0.0750, -1.0555, -0.0706], [-0.1946, -0.8593, -0.2238]], device='cuda:0')
  • 29.
    Example October 9, 2018 28 •Tensor #from_numpy, numpy import numpy as np a = np.array([3.5, 4]) b = tc.from_numpy(a) c = b.numpy() array a : [3.5 4. ] Tensor b : tensor([ 3.5000, 4.0000], dtype=torch.float64) array c : [3.5 4. ]
  • 30.
    Example October 9, 2018 29 •Datasets The following datasets are available: MNIST Fashion-MNIST EMNIST COCO LSUN ImageFolder DatasetFolder Imagenet-12 CIFAR STL10 SVHN PhotoTour : 바로 사용 가능. : 추가적인 과정 필요. : Custom dataset에 활용. MNIST Fashion-MNIST CIFAR LSUN CelebA K-pop(Custom)
  • 31.
    Example October 9, 2018 30 •Datasets – MNIST trans  Compose()is used when there are multiple transform options. Here, ToTensor() and Normalize(mean, std) are used.  ToTensor () changes the PIL Image to a tensor. torchvision dataset The default type is PIL Image.  Normalize (mean, std) transforms the range of the image. Here, the value of [0, 1] is adjusted to [-1, 1]. ((value-mean) / std) trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) import torch as tc import torchvision as tv import torchvision.transforms as transforms
  • 32.
    Example October 9, 2018 31 •Datasets – MNIST dataset  root : This is the path to store (MNIST data). Folders are automatically created with the specified name.  train : Set the data to be used for the train.  transform : Transform the data according to the transform option set previously.  download : Download (MINST data). (If you downloaded it once, it will not do it again.) trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) import torch as tc import torchvision as tv import torchvision.transforms as transforms
  • 33.
    Example October 9, 2018 32 •Datasets – MNIST dataloader  dataset : Set the dataset to load.  batch_size : Set the batch size.  shuffle : Shuffle the data and load it. trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) import torch as tc import torchvision as tv import torchvision.transforms as transforms
  • 34.
    Example October 9, 2018 33 •Datasets – LSUN https://github.com/fyu/lsun
  • 35.
    Example October 9, 2018 34 •Datasets – LSUN Run download.py def list_categories(tag): url = 'http://lsun.cs.princeton.edu/htbin/list.cgi?tag=' + tag f = urlopen(url) return json.loads(f.read()) If you are using Python 3.0 or later, modify the code from urllib2.urlopen (url) to urlopen (url). download.py -c bedroom bedroom 0 bridge 1 church_outdoor 2 classroom 3 conference_room 4 dining_room 5 kitchen 6 living_room 7 restaurant 8 tower 9
  • 36.
    Example October 9, 2018 35 •Datasets – LSUN trans = transforms.Compose([transforms.Resize((64,64)) ,transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.LSUN('.', classes=['bedroom_train'], transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) trans  Resize() is used to resize the image. datasets  root : Root directory for the database files.  classes : One of {‘train’, ‘val’, ‘test’} or a list of categories to load. e,g. [‘bedroom_train’, ‘church_train’].
  • 37.
    Example October 9, 2018 36 •Datasets – ImageFolder(CelebA) http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html
  • 38.
    Example October 9, 2018 37 •Datasets – ImageFolder(CelebA) http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html 압축을 풀면 jpg 형태로 파일들이 저장된 것을 확인할 수 있음.
  • 39.
    Example October 9, 2018 38 •Datasets – ImageFolder(CelebA) trans = transforms.Compose([transforms.Resize((64,64)) ,transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.ImageFolder('./img_align_celeba', trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) datasets  root : Root directory for the database files.  transform : Transform the data according to the transform option set previously.
  • 40.
    H Example October 9, 2018 39 •Neural Nets Sequence baseClass base model = tc.nn.Sequential( tc.nn.Linear(D_in, H), tc.nn.ReLU(), tc.nn.Linear(H, D_out), tc.nn.Sigmoid() ) class Model(tc.nn.Module): def __init__(self): super(Model, self).__init__() self.linear1 = tc.nn.Linear(D_in, H), self.linear2 = tc.nn.Linear(H, D_out) def forward(self, input): x = tc.nn.functional.relu(self.linear1(input)) x = tc.nn.functional.sigmoid(self.linear2(x)) return x D_in D_out Data manipulation
  • 41.
    Example October 9, 2018 40 •Neural Nets Class base class Model(tc.nn.Module): def __init__(self): super(Model, self).__init__() self.linear1 = tc.nn.Linear(D_in, H), self.linear2 = tc.nn.Linear(H, D_out) def forward(self, input): x = tc.nn.functional.relu(self.linear1(input)) x = tc.nn.functional.sigmoid(self.linear2(x)) return x tc.nn.Module functions Model functions 생성자 부모 클래스 초기화 Propagation Cf. Diamond problem A B C D Class D __init__() Class B __init__() Class A __init__() Class C __init__() Class A __init__()
  • 42.
    Example October 9, 2018 41 •Learning Back Propagation (Learning) Loss Optimizer loss_func = tc.nn.MSELoss() opt = tc.optim.Adam(model.parameters(), lr=0.01 ) for ep in range(epoch_sz): for step, (images, labels) in enumerate(dataloader): opt.zero_grad() images = images.to(device) labels = labels.to(device) output = model(images) loss = loss_func(output) loss.backward() opt.step()
  • 43.
    Example October 9, 2018 42 •Learning loss_func = tc.nn.MSELoss() opt = tc.optim.Adam(model.parameters(), lr=0.01 ) for ep in range(epoch_sz): for step, (images, labels) in enumerate(dataloader): opt.zero_grad() images = images.to(device) labels = labels.to(device) output = model(images) loss = loss_func(output) loss.backward() opt.step() Loss 선언 Optimizer 선언 (업데이트 하려는 parameter) 정해진 epoch 만큼 수행 앞서 설정한 dataloader에 따라 image(data)와 label(_) 불러옴 Gradient 초기화. Loss 구함 Back propagation(Gradient 구함) 정해진 optimizer에 따라 parameter 업데이트 trans = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]) datasets = tv.datasets.MNIST(root='./MNIST', train=True, download=True, transform=trans) dataloader = tc.utils.data.DataLoader(datasets=datasets, batch_size=100, shuffle=True) Cf.
  • 44.
    Example October 9, 2018 43 •Learning • L1Loss • MSELoss • CrossEntropyLoss • NLLLoss • PoissonNLLLoss • KLDivLoss • BCELoss • BCEWithLogitsLoss • MarginRankingLoss • HingeEmbeddingLoss • SmoothL1Loss • SoftMarginLoss • MultiLabelSoftMarginLoss • CosineEmbeddingLoss • MultiMarginLoss • TripleMarginLoss • Adadelta • Adagrad • Adam • SparseAdam • Adamax • ASGD • LBFGS • RMSprop • Rprop • SGD Loss Optimizer * Bold체는 주로 사용하는 함수들.
  • 45.
    Example • Applications –VanillaGAN 2018-10-09 44 https://github.com/messy-snail/GAN_PyTorch “What is this? Gum? It's GAN.” Cf.
  • 46.
    Example • Applications –VanillaGAN 2018-10-09 45 Sequence baseClass base Neural Nets
  • 47.
  • 48.
    Example • Applications –VanillaGAN 2018-10-09 47 Images(batch=100) … 28 28 100 Reshape(28*28*1->784) 7 8 4 7 8 4… 7 8 4 100 1 0 0 Noise(z) Label 1 0 Real Fake
  • 49.
    Example • Applications –VanillaGAN 2018-10-09 48 1 0 0 Noise(z) Label 1 0 Real Fake G (z) Generate D(Fakes) D(Images) Discriminate Loss Real Loss Fake Loss Reshape(28*28*1->784) 7 8 4 7 8 4… 7 8 4 100
  • 50.
    Example • Applications –VanillaGAN 2018-10-09 49 D(Fakes) D(Images) Discriminate Loss Real Loss Fake Total Loss Loss Update
  • 51.
    Example • Applications –VanillaGAN 2018-10-09 50 1 0 0 Noise(z) Label 1 Real G (z) Generate D(Fakes) Discriminate Loss g Loss Update
  • 52.
    Future Work October 9,2018 51 Paper Review Vanilla GAN DCGAN InfoGAN Unrolled GAN Wasserstein GAN LS GAN BEGAN Pix2Pix Cycle GAN Proposed Model SpyGAN Tools Document Programming PyTorch Python executable & UI Mathematical Study Linear algebra Probability and statistics Information theory Others Level Processor Ice Propagation Maybe next seminar?
  • 53.