TengdaHan

Multi-node-training on slurm with PyTorch

What's this?

• A simple note for how to start multi-node-training on slurm scheduler with PyTorch.
• Useful especially when scheduler is too busy that you cannot get multiple GPUs allocated, or you need more than 4 GPUs for a single job.
• Requirement: Have to use PyTorch DistributedDataParallel(DDP) for this purpose.
• Warning: might need to re-factor your own code.
• Warning: might be secretly condemned by your colleagues because using too many GPUs.

jinglescode

import torch
import torch.nn as nn

class conv_block_nested(nn.Module):

    def __init__(self, in_ch, mid_ch, out_ch):
        super(conv_block_nested, self).__init__()
        self.activation = nn.ReLU(inplace=True)
        self.conv1 = nn.Conv2d(in_ch, mid_ch, kernel_size=3, padding=1, bias=True)
        self.bn1 = nn.BatchNorm2d(mid_ch)

jeremyjordan

def soft_dice_loss(y_true, y_pred, epsilon=1e-6): 
    ''' 
    Soft dice loss calculation for arbitrary batch size, number of classes, and number of spatial dimensions.
    Assumes the `channels_last` format.
  
    # Arguments
        y_true: b x X x Y( x Z...) x c One hot encoding of ground truth
        y_pred: b x X x Y( x Z...) x c Network output, must sum to 1 over c channel (such as after softmax) 
        epsilon: Used for numerical stability to avoid divide by zero errors
    

bmabey

"""
Functions that implement some of the same functionality found in Matlab's bwmorph.

`thin` - was taken and adapted from https://gist.github.com/joefutrelle/562f25bbcf20691217b8
`spur` - Not perfect but pretty close to what matlab does via LUTs
`endpoints` - lines up perfectly with matlab's output (in my limited testing)
`branches` - this results in more clustered pixels than matlab's version but it pretty close
"""
import numpy as np
import scipy.ndimage as ndi

nasimrahaman

import torch
import torch.nn as nn


def log_sum_exp(x):
    # See implementation detail in
    # http://timvieira.github.io/blog/post/2014/02/11/exp-normalize-trick/
    # b is a shift factor. see link.
    # x.size() = [N, C]:
    b, _ = torch.max(x, 1)

chsasank

import numpy as np
from scipy.ndimage.interpolation import map_coordinates
from scipy.ndimage.filters import gaussian_filter

def elastic_transform(image, alpha, sigma, random_state=None):
    """Elastic deformation of images as described in [Simard2003]_.


    .. [Simard2003] Simard, Steinkraus and Platt, "Best Practices for
       Convolutional Neural Networks applied to Visual Document Analysis", in

fmder

import numpy
from scipy.ndimage.interpolation import map_coordinates
from scipy.ndimage.filters import gaussian_filter

def elastic_transform(image, alpha, sigma, random_state=None):
    """Elastic deformation of images as described in [Simard2003]_.


    .. [Simard2003] Simard, Steinkraus and Platt, "Best Practices for
       Convolutional Neural Networks applied to Visual Document Analysis", in

guweigang

git init                                                  # 初始化本地git仓库（创建新仓库）
git config --global user.name "xxx"                       # 配置用户名
git config --global user.email "xxx@xxx.com"              # 配置邮件
git config --global color.ui true                         # git status等命令自动着色
git config --global color.status auto
git config --global color.diff auto
git config --global color.branch auto
git config --global color.interactive auto
git config --global --unset http.proxy                    # remove  proxy configuration on git
git clone git+ssh://git@192.168.53.168/VT.git             # clone远程仓库