标签归档:colab

如何防止Google Colab断开连接?

问题:如何防止Google Colab断开连接?

问:是否可以通过编程方式防止Google Colab在超时时断开连接?

下面介绍导致笔记本计算机自动断开连接的情况:

Google Colab笔记本的空闲超时为90分钟,绝对超时为12小时。这意味着,如果用户在超过90分钟的时间内未与其Google Colab笔记本互动,则其实例将自动终止。另外,Colab实例的最大生存期为12小时。

自然,我们希望自动将最大值从实例中挤出,而不必不断地手动与之交互。在这里,我将假定常见的系统要求:

  • Ubuntu 18 LTS / Windows 10 / Mac操作系统
  • 对于基于Linux的系统,请使用流行的DE,例如Gnome 3或Unity
  • Firefox或Chromium浏览器

我要在这里指出,这种行为并未违反 Google Colab的使用条款,尽管根据其常见问题解答不鼓励这样做(简而言之:从道德上讲,如果您真的不需要它,则用尽所有GPU是不可行的))。


我当前的解决方案非常愚蠢:

  • 首先,我关闭屏幕保护程序,因此我的屏幕始终保持打开状态。
  • 我有一个Arduino开发板,所以我只是将它变成了一个橡胶鸭子USB,并使其在我睡觉时模拟原始用户交互(只是因为我手边有其他用例)。

有更好的方法吗?

Q: Is there any way to programmatically prevent Google Colab from disconnecting on a timeout?

The following describes the conditions causing a notebook to automatically disconnect:

Google Colab notebooks have an idle timeout of 90 minutes and absolute timeout of 12 hours. This means, if user does not interact with his Google Colab notebook for more than 90 minutes, its instance is automatically terminated. Also, maximum lifetime of a Colab instance is 12 hours.

Naturally, we want to automatically squeeze the maximum out of the instance, without having to manually interact with it constantly. Here I will assume commonly seen system requirements:

  • Ubuntu 18 LTS / Windows 10 / Mac Operating systems
  • In case of Linux-based systems, using popular DEs like Gnome 3 or Unity
  • Firefox or Chromium browsers

I should point out here that such behavior does not violate Google Colab’s Terms of Use, although it is not encouraged according to their FAQ (in short: morally it is not okay to use up all of the GPUs if you don’t really need it).


My current solution is very dumb:

  • First, I turn the screensaver off, so my sreen is always on.
  • I have an Arduino board, so I just turned it into a rubber ducky usb and make it emulate primitive user interaction while I sleep (just because I have it at hand for other use-cases).

Are there better ways?


回答 0

编辑: 显然,该解决方案非常简单,并且不需要任何JavaScript。只需在底部创建具有以下行的新单元格:

while True:pass

现在将单元格保持在运行顺序中,以便无限循环不会停止,从而使会话保持活动状态。

旧方法: 设置一个JavaScript间隔,每60秒点击一次connect按钮。使用Ctrl + Shift + I打开开发人员设置(在您的Web浏览器中),然后单击控制台选项卡,然后在控制台提示符下键入此设置。(对于Mac,请按Option + Command + I)

function ConnectButton(){
    console.log("Connect pushed"); 
    document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click() 
}
setInterval(ConnectButton,60000);

Edit: Apparently the solution is very easy, and doesn’t need any JavaScript. Just create a new cell at the bottom having the following line:

while True:pass

now keep the cell in the run sequence so that the infinite loop won’t stop and thus keep your session alive.

Old method: Set a javascript interval to click on the connect button every 60 seconds. Open developer-settings (in your web-browser) with Ctrl+Shift+I then click on console tab and type this on the console prompt. (for mac press Option+Command+I)

function ConnectButton(){
    console.log("Connect pushed"); 
    document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click() 
}
setInterval(ConnectButton,60000);

回答 1

由于现在将连接按钮的ID更改为“ colab-connect-button”,因此可以使用以下代码来继续单击该按钮。

function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("colab-connect-button").click()
}
setInterval(ClickConnect,60000)

如果仍然无法解决问题,请按照以下步骤操作:

  1. 右键单击连接按钮(位于colab的右上方)
  2. 点击检查
  3. 获取按钮的HTML ID并替换为以下代码
function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("Put ID here").click() // Change id here
}
setInterval(ClickConnect,60000)

Since the id of the connect button is now changed to “colab-connect-button”, the following code can be used to keep clicking on the button.

function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("colab-connect-button").click()
}
setInterval(ClickConnect,60000)

If still, this doesn’t work, then follow the steps given below:

  1. Right-click on the connect button (on the top-right side of the colab)
  2. Click on inspect
  3. Get the HTML id of the button and substitute in the following code
function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("Put ID here").click() // Change id here
}
setInterval(ClickConnect,60000)

回答 2

嗯,这对我有用-

在控制台中运行以下代码,它将阻止您断开连接。Ctrl + Shift + i打开检查器视图。然后进入控制台。

function ClickConnect(){
    console.log("Working"); 
    document.querySelector("colab-toolbar-button#connect").click() 
}
setInterval(ClickConnect,60000)

如何防止Google Colab断开连接

Well this is working for me –

run the following code in the console and it will prevent you from disconnecting. Ctrl+ Shift + i to open inspector view . Then go to console.

function ClickConnect(){
    console.log("Working"); 
    document.querySelector("colab-toolbar-button#connect").click() 
}
setInterval(ClickConnect,60000)

How to prevent google colab from disconnecting


回答 3

对我而言,以下示例:

  • document.querySelector("#connect").click() 要么
  • document.querySelector("colab-toolbar-button#connect").click() 要么
  • document.querySelector("colab-connect-button").click()

抛出错误。

我必须使它们适应以下条件:

版本1:

function ClickConnect(){
  console.log("Connnect Clicked - Start"); 
  document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click();
  console.log("Connnect Clicked - End"); 
};
setInterval(ClickConnect, 60000)

版本2: 如果您希望能够停止该功能,请使用以下新代码:

var startClickConnect = function startClickConnect(){
    var clickConnect = function clickConnect(){
        console.log("Connnect Clicked - Start");
        document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click();
        console.log("Connnect Clicked - End"); 
    };

    var intervalId = setInterval(clickConnect, 60000);

    var stopClickConnectHandler = function stopClickConnect() {
        console.log("Connnect Clicked Stopped - Start");
        clearInterval(intervalId);
        console.log("Connnect Clicked Stopped - End");
    };

    return stopClickConnectHandler;
};

var stopClickConnect = startClickConnect();

为了停止,请调用:

stopClickConnect();

For me the following examples:

  • document.querySelector("#connect").click() or
  • document.querySelector("colab-toolbar-button#connect").click() or
  • document.querySelector("colab-connect-button").click()

were throwing errors.

I had to adapt them to the following:

Version 1:

function ClickConnect(){
  console.log("Connnect Clicked - Start"); 
  document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click();
  console.log("Connnect Clicked - End"); 
};
setInterval(ClickConnect, 60000)

Version 2: If you would like to be able to stop the function, here is the new code:

var startClickConnect = function startClickConnect(){
    var clickConnect = function clickConnect(){
        console.log("Connnect Clicked - Start");
        document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click();
        console.log("Connnect Clicked - End"); 
    };

    var intervalId = setInterval(clickConnect, 60000);

    var stopClickConnectHandler = function stopClickConnect() {
        console.log("Connnect Clicked Stopped - Start");
        clearInterval(intervalId);
        console.log("Connnect Clicked Stopped - End");
    };

    return stopClickConnectHandler;
};

var stopClickConnect = startClickConnect();

In order to stop, call:

stopClickConnect();

回答 4

使用Pynput在您的PC中创建python代码

from pynput.mouse import Button, Controller
import time

mouse = Controller()

while True:
    mouse.click(Button.left, 1)
    time.sleep(30)

在您的桌面上运行此代码,然后将鼠标指针悬停在任何目录的目录结构上(左侧的左侧栏-文件部分),此代码将每30秒不断单击一次目录,因此每30秒将展开和缩小一次,因此您的会话不会过期重要-您必须在PC中运行此代码

create a python code in your pc with pynput

from pynput.mouse import Button, Controller
import time

mouse = Controller()

while True:
    mouse.click(Button.left, 1)
    time.sleep(30)

Run this code in your Desktop, Then point mouse arrow over (colabs left panel – file section) directory structure on any directory this code will keep clicking on directory on every 30 seconds so it will expand and shrink every 30 seconds so your session will not get expired Important – you have to run this code in your pc


回答 5

我没有单击“连接”按钮,而是单击“评论”按钮以使会话保持活动状态。(2020年8月)

function ClickConnect(){

console.log("Working"); 
document.querySelector("#comments > span").click() 
}
setInterval(ClickConnect,5000)

Instead of clicking the connect button, i just clicking on comment button to keep my session alive. (August-2020)

function ClickConnect(){

console.log("Working"); 
document.querySelector("#comments > span").click() 
}
setInterval(ClickConnect,5000)

回答 6

我使用宏程序定期单击RAM / Disk按钮以整夜训练模型。诀窍是配置一个宏程序,以两次单击Ram / Disk Colab工具栏按钮,两次单击之间的间隔很短,这样即使运行时断开连接,它也将重新连接。(第一次单击用于关闭对话框,第二次单击用于重新连接)。但是,您仍然必须整夜打开笔记本电脑,甚至可以固定Colab标签。

I use a Macro Program to periodically click on the RAM/Disk button to train the model all night. The trick is to configure a macro program to click on the Ram/Disk Colab Toolbar Button twice with a short interval between the two clicks so that even if the Runtime gets disconnected it will reconnect back. (the first click used to close the dialog box and the second click used to RECONNECT). However, you still have to leave your laptop open all night and maybe pin the Colab tab.


回答 7

在某些脚本的帮助下,以上答案可能效果很好。对于没有脚本的烦人的断开连接,我有一个解决方案(或一种技巧),尤其是当您的程序必须从google驱动器读取数据时,例如训练深度学习网络模型时,使用脚本进行reconnect操作就没有用了,因为一旦您断开与colab的连接,该程序就死了,应该再次手动连接到Google驱动器,以使您的模型能够再次读取数据集,但是脚本不会执行此操作。
我已经测试了很多次,并且效果很好。
当您使用浏览器(我使用Chrome)在colab页面上运行程序时,请记住,一旦程序开始运行,就不要对浏览器进行任何操作,例如:切换到其他网页,打开或关闭另一个网页,以及依此类推,只需将其放置在那里,等待程序完成运行,就可以切换到pycharm等其他软件来继续编写代码,而不必切换到另一个网页。我不知道为什么打开或关闭或切换到其他页面会导致google colab页面的连接问题,但是每次我尝试打扰我的浏览器(如执行某些搜索工作)时,我与colab的连接都会很快断开。

The above answers with the help of some scripts maybe work well. I have a solution(or a kind of trick) for that annoying disconnection without scripts, especially when your program must read data from your google drive, like training a deep learning network model, where using scripts to do reconnect operation is of no use because once you disconnect with your colab, the program is just dead, you should manually connect to your google drive again to make your model able to read dataset again, but the scripts will not do that thing.
I’ve already test it many times and it works well.
When you run a program on the colab page with a browser(I use Chrome), just remember that don’t do any operation to your browser once your program starts running, like: switch to other webpages, open or close another webpage, and so on, just just leave it alone there and waiting for your program finish running, you can switch to another software, like pycharm to keep writing your codes but not switch to another webpage. I don’t know why open or close or switch to other pages will cause the connection problem of the google colab page, but each time I try to bothered my browser, like do some search job, my connection to colab will soon break down.


回答 8

尝试这个:

function ClickConnect(){
  console.log("Working"); 
  document
    .querySelector("#top-toolbar > colab-connect-button")
    .shadowRoot
    .querySelector("#connect")
    .click()
}

setInterval(ClickConnect,60000)

Try this:

function ClickConnect(){
  console.log("Working"); 
  document
    .querySelector("#top-toolbar > colab-connect-button")
    .shadowRoot
    .querySelector("#connect")
    .click()
}

setInterval(ClickConnect,60000)

回答 9

使用python硒

from selenium.webdriver.common.keys import Keys
from selenium import webdriver
import time   

driver = webdriver.Chrome('/usr/lib/chromium-browser/chromedriver')

notebook_url = ''
driver.get(notebook_url)

# run all cells
driver.find_element_by_tag_name('body').send_keys(Keys.CONTROL + Keys.F9)
time.sleep(5)

# click to stay connected
start_time = time.time()
current_time = time.time()
max_time = 11*59*60 #12hours

while (current_time - start_time) < max_time:
    webdriver.ActionChains(driver).send_keys(Keys.ESCAPE).perform()
    driver.find_element_by_xpath('//*[@id="top-toolbar"]/colab-connect-button').click()
    time.sleep(30)
    current_time = time.time()

Using python selenium

from selenium.webdriver.common.keys import Keys
from selenium import webdriver
import time   

driver = webdriver.Chrome('/usr/lib/chromium-browser/chromedriver')

notebook_url = ''
driver.get(notebook_url)

# run all cells
driver.find_element_by_tag_name('body').send_keys(Keys.CONTROL + Keys.F9)
time.sleep(5)

# click to stay connected
start_time = time.time()
current_time = time.time()
max_time = 11*59*60 #12hours

while (current_time - start_time) < max_time:
    webdriver.ActionChains(driver).send_keys(Keys.ESCAPE).perform()
    driver.find_element_by_xpath('//*[@id="top-toolbar"]/colab-connect-button').click()
    time.sleep(30)
    current_time = time.time()

回答 10

我认为JavaScript解决方案不再有效。我在笔记本中使用以下命令进行操作:

    from IPython.display import display, HTML
    js = ('<script>function ConnectButton(){ '
           'console.log("Connect pushed"); '
           'document.querySelector("#connect").click()} '
           'setInterval(ConnectButton,3000);</script>')
    display(HTML(js))

首次执行全部运行时(在启动JavaScript或Python代码之前),控制台将显示:

Connected to 
wss://colab.research.google.com/api/kernels/0e1ce105-0127-4758-90e48cf801ce01a3/channels?session_id=5d8...

但是,每次运行JavaScript时,您都会看到console.log部分,但是click部分仅给出:

Connect pushed

Uncaught TypeError: Cannot read property 'click' of null
 at ConnectButton (<anonymous>:1:92)

其他人建议将按钮名称更改为#colab-connect-button,但这会产生相同的错误。

启动运行系统后,该按钮将更改为显示RAM / DISK,并显示一个下拉列表。单击下拉列表创建一个<DIV class=goog menu...>以前未在DOM中显示的新内容,并带有2个选项“连接到托管运行时”和“连接到本地运行时”。如果控制台窗口已打开并显示元素,则在单击下拉元素时可以看到此DIV出现。只需在出现的新窗口中的两个选项之间移动鼠标焦点,即可向DOM添加其他元素,一旦鼠标释放焦点,它们便会从DOM中完全删除,甚至无需单击即可。

I don’t believe the JavaScript solutions work anymore. I was doing it from within my notebook with:

    from IPython.display import display, HTML
    js = ('<script>function ConnectButton(){ '
           'console.log("Connect pushed"); '
           'document.querySelector("#connect").click()} '
           'setInterval(ConnectButton,3000);</script>')
    display(HTML(js))

When you first do a Run all (before the JavaScript or Python code has started), the console displays:

Connected to 
wss://colab.research.google.com/api/kernels/0e1ce105-0127-4758-90e48cf801ce01a3/channels?session_id=5d8...

However, ever time the JavaScript runs, you see the console.log portion, but the click portion simply gives:

Connect pushed

Uncaught TypeError: Cannot read property 'click' of null
 at ConnectButton (<anonymous>:1:92)

Others suggested the button name has changed to #colab-connect-button, but that gives same error.

After the runtime is started, the button is changed to show RAM/DISK, and a drop down is presented. Clicking on the drop down creates a new <DIV class=goog menu...> that was not shown in the DOM previously, with 2 options “Connect to hosted runtime” and “Connect to local runtime”. If the console window is open and showing elements, you can see this DIV appear when you click the dropdown element. Simply moving the mouse focus between the two options in the new window that appears adds additional elements to the DOM, as soon as the mouse looses focus, they are removed from the DOM completely, even without clicking.


回答 11

我尝试了上面的代码,但它们对我不起作用。这是我重新连接的JS代码。

let interval = setInterval(function(){
let ok = document.getElementById('ok');
if(ok != null){
   console.log("Connect pushed");
ok.click();
}},60000)

您可以使用相同的方式(在浏览器的控制台上运行)来运行它。如果要停止脚本,可以输入clearInterval(interval)并再次运行setInterval(interval)

我希望这可以帮助你。

I tried the codes above but they did not work for me. So here is my JS code for reconnecting.

let interval = setInterval(function(){
let ok = document.getElementById('ok');
if(ok != null){
   console.log("Connect pushed");
ok.click();
}},60000)

You can use it with the same way (run it on the console of your browser) to run it. If you want to stop the script, you can enter clearInterval(interval) and want to run again setInterval(interval).

I hope this helps you.


回答 12

更新了一个。这个对我有用。

function ClickConnect(){
console.log("Working"); 
document.querySelector("paper-icon-button").click()
}
Const myjob = setInterval(ClickConnect, 60000)

如果对您不起作用,请尝试运行以下命令清除它:

clearInterval(myjob)

Updated one. it works for me.

function ClickConnect(){
console.log("Working"); 
document.querySelector("paper-icon-button").click()
}
Const myjob = setInterval(ClickConnect, 60000)

If isn’t working you for you guys try clear it by running:

clearInterval(myjob)

回答 13

这对我有用(似乎他们更改了按钮的类名或ID):

function ClickConnect(){
    console.log("Working"); 
    document.querySelector("colab-connect-button").click() 
}
setInterval(ClickConnect,60000)

This one worked for me (it seems like they changed the button classname or id) :

function ClickConnect(){
    console.log("Working"); 
    document.querySelector("colab-connect-button").click() 
}
setInterval(ClickConnect,60000)

回答 14

投票得最多的答案当然对我有用,但这会使“管理会话”窗口一次又一次地弹出。
我已经解决了这一问题,方法是使用浏览器控制台如下所示自动单击刷新按钮

function ClickRefresh(){
    console.log("Clicked on refresh button"); 
    document.querySelector("paper-icon-button").click()
}
setInterval(ClickRefresh, 60000)

随时在此要点上为此贡献更多代码片段https://gist.github.com/Subangkar/fd1ef276fd40dc374a7c80acc247613e

The most voted answer certainly works for me but it makes the Manage session window popping up again and again.
I’ve solved that by auto clicking the refresh button using browser console like below

function ClickRefresh(){
    console.log("Clicked on refresh button"); 
    document.querySelector("paper-icon-button").click()
}
setInterval(ClickRefresh, 60000)

Feel free to contribute more snippets for this at this gist https://gist.github.com/Subangkar/fd1ef276fd40dc374a7c80acc247613e


回答 15

也许以前的许多解决方案都不再起作用。例如,下面的代码继续在Colab中创建新的代码单元,但仍在工作。无疑,创建一堆代码单元是一个不便之处。如果在运行几个小时后创建了太多的代码单元,而没有足够的RAM,则浏览器可能会冻结。

反复创建代码单元-

function ClickConnect(){
console.log("Working"); 
document.querySelector("colab-toolbar-button").click() 
}setInterval(ClickConnect,60000)

但是我发现下面的代码正在运行,它不会引起任何问题。在Colab笔记本选项卡中,Ctrl + Shift + i同时单击该键,然后将以下代码粘贴到控制台中。120000个间隔就足够了。

function ClickConnect(){
console.log("Working"); 
document.querySelector("colab-toolbar-button#connect").click() 
}setInterval(ClickConnect,120000)

我已在2020年11月在firefox中测试了此代码。它也将在chrome上工作。

Perhaps many of the previous solutions are no longer working. For example, this bellow code continues to create new code cells in Colab, working though. Undoubtedly, creating a bunch of code cells is an inconvenience. If too many code cells are created in some hours of running and there is no enough RAM, the browser may freeze.

This repetedly creates code cells—

function ClickConnect(){
console.log("Working"); 
document.querySelector("colab-toolbar-button").click() 
}setInterval(ClickConnect,60000)

But I found the code below is working, it doesn’t cause any problems. In the Colab notebook tab, click on the Ctrl + Shift + i key simultaneously and paste the below code in the console. 120000 intervals are enough.

function ClickConnect(){
console.log("Working"); 
document.querySelector("colab-toolbar-button#connect").click() 
}setInterval(ClickConnect,120000)

I have tested this code in firefox, in November 2020. It will work on chrome too.


回答 16

我建议使用JQuery(似乎Co-lab默认包含JQuery)。

function ClickConnect(){
  console.log("Working");
  $("colab-toolbar-button").click();
}
setInterval(ClickConnect,60000);

I would recommend using JQuery (It seems that Co-lab includes JQuery by default).

function ClickConnect(){
  console.log("Working");
  $("colab-toolbar-button").click();
}
setInterval(ClickConnect,60000);

回答 17

这些JavaScript函数存在问题:

function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("colab-connect-button").click()
}
setInterval(ClickConnect,60000)

他们实际在单击按钮之前在控制台上打印“ Clicked on connect button”。从该线程的不同答案中可以看出,自Google Colab启动以来,connect按钮的ID已经更改了两次。而且将来也可能会更改。因此,如果您打算从该线程中复制旧答案,则可能会说“单击了连接按钮”,但实际上可能不会这样做。当然,如果单击不起作用,它将在控制台上显示一个错误,但是如果您可能不会意外看到该怎么办?因此,您最好这样做:

function ClickConnect(){
    document.querySelector("colab-connect-button").click()
    console.log("Clicked on connect button"); 
}
setInterval(ClickConnect,60000)

您肯定会看到它是否真正起作用。

I have a problem with these javascript functions:

function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("colab-connect-button").click()
}
setInterval(ClickConnect,60000)

They print the “Clicked on connect button” on the console before the button is actually clicked. As you can see from different answers in this thread, the id of the connect button has changed a couple of times since Google Colab was launched. And it could be changed in the future as well. So if you’re going to copy an old answer from this thread it may say “Clicked on connect button” but it may actually not do that. Of course if the clicking won’t work it will print an error on the console but what if you may not accidentally see it? So you better do this:

function ClickConnect(){
    document.querySelector("colab-connect-button").click()
    console.log("Clicked on connect button"); 
}
setInterval(ClickConnect,60000)

And you’ll definitely see if it truly works or not.


回答 18

function ClickConnect()
{
    console.log("Working...."); 
    document.querySelector("paper-button#comments").click()
}
setInterval(ClickConnect,600)

这对我有用,但明智地使用

快乐学习:)

function ClickConnect()
{
    console.log("Working...."); 
    document.querySelector("paper-button#comments").click()
}
setInterval(ClickConnect,600)

this worked for me but use wisely

happy learning :)


回答 19

以下最新解决方案适用于我:

function ClickConnect(){
  colab.config
  console.log("Connnect Clicked - Start"); 
  document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click();
  console.log("Connnect Clicked - End");
};
setInterval(ClickConnect, 60000)

the following LATEST solution works for me:

function ClickConnect(){
  colab.config
  console.log("Connnect Clicked - Start"); 
  document.querySelector("#top-toolbar > colab-connect-button").shadowRoot.querySelector("#connect").click();
  console.log("Connnect Clicked - End");
};
setInterval(ClickConnect, 60000)

回答 20

下面的javascript对我有用。学分@ artur.k.space

function ColabReconnect() {
    var dialog = document.querySelector("colab-dialog.yes-no-dialog");
    var dialogTitle = dialog && dialog.querySelector("div.content-area>h2");
    if (dialogTitle && dialogTitle.innerText == "Runtime disconnected") {
        dialog.querySelector("paper-button#ok").click();
        console.log("Reconnecting...");
    } else {
        console.log("ColabReconnect is in service.");
    }
}
timerId = setInterval(ColabReconnect, 60000);

在Colab笔记本中,同时单击Ctrl + Shift +i键。将脚本复制并粘贴到提示行中。然后Enter在关闭编辑器之前点击。

这样,该功能将每60秒检查一次,以查看是否显示了屏幕连接对话框,如果显示,则该功能将ok自动为您单击该按钮。

The javascript below works for me. Credits to @artur.k.space.

function ColabReconnect() {
    var dialog = document.querySelector("colab-dialog.yes-no-dialog");
    var dialogTitle = dialog && dialog.querySelector("div.content-area>h2");
    if (dialogTitle && dialogTitle.innerText == "Runtime disconnected") {
        dialog.querySelector("paper-button#ok").click();
        console.log("Reconnecting...");
    } else {
        console.log("ColabReconnect is in service.");
    }
}
timerId = setInterval(ColabReconnect, 60000);

In the Colab notebook, click on Ctrl + Shift + the i key simultaneously. Copy and paste the script into the prompt line. Then hit Enter before closing the editor.

By doing so, the function will check every 60 seconds to see if the onscreen connection dialog is shown, and if it is, the function would then click the ok button automatically for you.


回答 21

好吧,我不是python家伙,也不知道这个’Colab’的实际用途是什么,我将其用作构建系统。我以前在其中设置了ssh转发,然后将这段代码放到运行中,是的。

import getpass
authtoken = getpass.getpass()

Well I am not a python guy nor I know what is the actual use of this ‘Colab’, I use it as a build system lol. And I used to setup ssh forwarding in it then put this code and just leave it running and yeah it works.

import getpass
authtoken = getpass.getpass()

回答 22

此代码在文件资源管理器窗格中单击“刷新文件夹”。

function ClickRefresh(){
  console.log("Working"); 
  document.querySelector("[icon='colab:folder-refresh']").click()
}
const myjob = setInterval(ClickRefresh, 60000)

This code keep clicking “Refresh folder” in the file explorer pane.

function ClickRefresh(){
  console.log("Working"); 
  document.querySelector("[icon='colab:folder-refresh']").click()
}
const myjob = setInterval(ClickRefresh, 60000)

回答 23

GNU Colab使您可以在Colaboratory实例之上运行标准的持久桌面环境。

实际上,它包含一种不让机器死掉的机制。

这是一个视频演示

GNU Colab lets you run a standard persistent desktop environment on top of a Colaboratory instance.

Indeed it contains a mechanism to not let machines die of idling.

Here’s a video demonstration.


回答 24

您也可以使用Python按下箭头键。我也在以下代码中添加了一些随机性。

from pyautogui import press, typewrite, hotkey
import time
from random import shuffle

array = ["left", "right", "up", "down"]

while True:
    shuffle(array)
    time.sleep(10)
    press(array[0])
    press(array[1])
    press(array[2])
    press(array[3])

You can also use Python to press the arrow keys. I added a little bit of randomness in the following code as well.

from pyautogui import press, typewrite, hotkey
import time
from random import shuffle

array = ["left", "right", "up", "down"]

while True:
    shuffle(array)
    time.sleep(10)
    press(array[0])
    press(array[1])
    press(array[2])
    press(array[3])

回答 25

只需在要运行的单元格之后运行以下代码,以防止数据丢失。

!python

同样要退出此模式,请写

exit()

Just run the code below after the cell you want to run to save from data loss.

!python

Also to exit from this mode, write

exit()

回答 26

我一直在寻找解决方案,直到找到一个Python3,该Python3总是在同一位置来回移动鼠标并单击,但这足以使Colab误以为我在笔记本电脑上很活跃并且没有断开连接。

import numpy as np
import time
import mouse
import threading

def move_mouse():
    while True:
        random_row = np.random.random_sample()*100
        random_col = np.random.random_sample()*10
        random_time = np.random.random_sample()*np.random.random_sample() * 100
        mouse.wheel(1000)
        mouse.wheel(-1000)
        mouse.move(random_row, random_col, absolute=False, duration=0.2)
        mouse.move(-random_row, -random_col, absolute=False, duration = 0.2)
        mouse.LEFT
        time.sleep(random_time)


x = threading.Thread(target=move_mouse)
x.start()

您需要安装所需的软件包:sudo -H pip3 install <package_name> 您只需要使用(在本地计算机中)运行它即可(sudo因为它可以控制鼠标)并且它应该可以工作,从而使您能够充分利用Colab的12h会话。

积分: 对于使用Colab(Pro)的用户:防止会话由于不活动而断开连接

I was looking for a solution until I found a Python3 that randomly moves the mouse back and forth and clicks, always on the same place, but that’s enough to fool Colab into thinking I’m active on the notebook and not disconnect.

import numpy as np
import time
import mouse
import threading

def move_mouse():
    while True:
        random_row = np.random.random_sample()*100
        random_col = np.random.random_sample()*10
        random_time = np.random.random_sample()*np.random.random_sample() * 100
        mouse.wheel(1000)
        mouse.wheel(-1000)
        mouse.move(random_row, random_col, absolute=False, duration=0.2)
        mouse.move(-random_row, -random_col, absolute=False, duration = 0.2)
        mouse.LEFT
        time.sleep(random_time)


x = threading.Thread(target=move_mouse)
x.start()

You need to install the needed packages: sudo -H pip3 install <package_name> You just need to run it (in your local machine) with sudo (as it takes control of the mouse) and it should work, allowing you to take full advantage of Colab’s 12h sessions.

Credits: For those using Colab (Pro): Preventing Session from disconnecting due to inactivity


回答 27

function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("connect").click() // Change id here
}
setInterval(ClickConnect,60000)

试试上面对我有用的代码:)

function ClickConnect(){
    console.log("Clicked on connect button"); 
    document.querySelector("connect").click() // Change id here
}
setInterval(ClickConnect,60000)

Try above code it worked for me:)


Fastai – 快速深度学习模块

欢迎光临Fastai

FAST AI使用现代最佳实践简化了快速准确的神经网络训练

正在安装

您可以在不安装的情况下使用Fastai,方法是使用Google Colab事实上,本文档的每一页还可以作为交互式笔记本使用-单击任何页面顶部的“Open in CoLab”将其打开(请确保将Colab运行时更改为“GPU”以使其快速运行!)请参阅上的fast.ai文档Using Colab了解更多信息

只要你运行的是Linux或Windows(不支持Mac),你就可以用conda在你自己的机器上安装Fastai(强烈推荐)。对于Windows,请参阅“在Windows上运行”了解重要说明

如果您使用的是miniconda(推荐)然后运行(请注意,如果您替换conda使用mamba安装过程将更快、更可靠):

conda install -c fastchan fastai

或者如果您正在使用Anaconda然后运行:

conda install -c fastchan fastai anaconda

要使用pip安装,请使用:pip install fastai如果使用pip安装,则应首先按照PyTorch安装PyTorchinstallation instructions

如果您计划自己开发Fastai,或者想要走在前列,您可以使用可编辑安装(如果这样做,您还应该使用的可编辑安装fastcore来配合它。)首先安装PyTorch,然后:

git clone https://github.com/fastai/fastai
pip install -e "fastai[dev]"

学习速度快

开始快速学习(和深度学习)的最好方法是阅读the book,并完成the free course

要了解使用FAST AI可能实现的功能,请查看Quick Start,它展示了如何使用大约5行代码来构建图像分类器、图像分割模型、文本情感模型、推荐系统和表格模型。对于每个应用程序,代码基本相同

通读一下Tutorials学习如何在您自己的数据集上训练您自己的模型。使用导航侧栏浏览FAST AI文档。这里记录了每个类、函数和方法

要了解图书馆的设计和动机,请阅读peer reviewed paper

关于FAST

Fastai是一个深度学习库,它为从业者提供高级组件,这些组件可以快速、轻松地在标准深度学习领域提供最先进的结果,并为研究人员提供可以混合和匹配的低级组件,以建立新的方法。它的目标是在易用性、灵活性或性能方面不做实质性妥协的情况下完成这两件事。这要归功于精心分层的体系结构,该体系结构以分离的抽象方式表达了许多深度学习和数据处理技术的公共底层模式。通过利用底层Python语言的动态性和PyTorch库的灵活性,可以简洁而清晰地表达这些抽象。FAST AI包括:

  • 一种新的Python类型调度系统和张量的语义类型层次结构
  • 一个GPU优化的计算机视觉库,可以在纯Python中扩展
  • 一种优化器,它将现代优化器的常见功能重构为两个基本部分,允许在4-5行代码中实现优化算法
  • 一种新颖的双向回调系统,可以访问数据、模型或优化器的任何部分,并在培训期间随时更改它们
  • 一种全新的数据挡路接口
  • 还有更多

FAST AI围绕两个主要设计目标进行组织:平易近人和快速高效,同时也是深度可破解和可配置的。它构建在提供可组合构建块的低级API层次结构之上。这样,想要重写高级API的一部分或添加特定行为以满足其需求的用户不必学习如何使用最低级别

从其他库迁移

从普通的PyTorch、Ignite或任何其他基于PyTorch的库迁移非常容易,甚至可以将Fastai与其他库结合使用。通常,您将能够使用所有现有的数据处理代码,但将能够减少培训所需的代码量,并更容易地利用现代最佳实践。以下是一些受欢迎的库中的迁移指南,可帮助您上路:

Windows支持

使用安装时mambaconda替换-c fastchan在使用的安装中-c pytorch -c nvidia -c fastai,因为Windows当前不支持Fastchan

由于Jupyter和Windows上的Python多处理问题,num_workersDataloader自动重置为0,以避免Jupyter挂起。这使得Windows上的Jupyter上的计算机视觉等任务比Linux上的要慢很多倍。如果从脚本使用FastAI,则不存在此限制

看见this example要充分利用Windows上的Fastai API,请执行以下操作

测试

要并行运行测试,请启动:

nbdev_test_nbsmake test

要通过所有测试,您需要安装以下可选依赖项:

pip install "sentencepiece<0.1.90" wandb tensorboard albumentations pydicom opencv-python scikit-image pyarrow kornia \
    catalyst captum neptune-cli

测试是使用nbdev,例如,请参阅的文档test_eq

贡献

克隆此存储库后,请运行nbdev_install_git_hooks在你的航站楼里。这将设置git挂钩,它会清理笔记本以删除笔记本中存储的无关内容(例如,您运行了哪些单元格),这会导致不必要的合并冲突。

在提交PR之前,请检查当地图书馆和笔记本是否匹配。剧本nbdev_diff_nbs我可以让你知道当地图书馆和笔记本有什么不同

  • 如果您对其中一个导出的单元格中的笔记本进行了更改,则可以使用以下命令将其导出到库nbdev_build_libmake fastai
  • 如果您对库进行了更改,则可以使用以下命令将其导出回笔记本nbdev_update_lib

码头集装箱

对于那些对这个项目的官方码头集装箱感兴趣的人,可以在here