rahul-khanna / ml_experiments_for_gummies Goto Github PK

• Find the phrase "IAM & admin" and hover over it
• Click on Quotas
• Click on the button under "Metric" and hit None
• Then type in the seach bar (in that same menu) "GPUs (all regions)"
• Now request one GPU from Google, you'll have to fill out some information, unfortunately. Do so though. A good reason to put down is "ML Practioner trying to practice deep learning techniques"
• Pause in your setup process until Google sends you back an approval email. This should take no more than one day.

• If there are warnings, don't worry about them
• If there are errors, I would suggest googling the errors

• This is one way that you can connect to your machine. However, to fully use your machine you're going to want to follow the steps below.

• if this fails please run the following command: mkdir .ssh
• then run: cd .ssh

• Name your key: google_cloud_key
• You don't have to set a passphrase for your key, but I do

• Makes your keys read only

• google_cloud_key
• google_cloud_key.pub

• hit i
• make room at the top of the file (if something else is already there) by hitting enter
• paste what you copied from your own computer into authorized_keys (ctrl+v)
• hit the esc key on your computer
• hit the : key on your computer
• type wq
• hit enter

• ssh-add ~/.ssh/google_cloud_key (the ~ command might not work on Windows, so please just write out the full path to your newly created .ssh folder––from step 7)
• ssh -i ~/.ssh/google_cloud_key your_google_cloud_user_name@your_machines_external_ip
  • your_google_cloud_user_name = the part of your email before the @ sign
  • your_machines_external_ip = the numbers and periods found on the google compute engine dashboard under the heading External IP

• ssh-add ~/.ssh/google_cloud_key (you don't always need to run this, but in case you see an error, run this and the error should go away)
• ssh -i ~/.ssh/google_cloud_key your_google_cloud_user_name@your_machines_external_ip

• Make a note as to where you've saved this unzipped folder.

• You have now created a virtual environment called clouds. This is important when it comes to replication of your work.

• You should now see the words (clouds) on your screen

• If you have connected to your instance via the SSH button on the Google Compute Engine Dashboard find the settings button in top right portion of your screen (the button that looks like a wheel). Find the upload file option, and upload the provided "rqs.txt" file

• If you have connected via your command prompt/terminal, open a new tab in your command prompt/terminal and navigate to where you saved the unzipped folder. Then run the following command:

  scp -i ~/.ssh/google_cloud_key rqs.text your_google_cloud_user_name@your_machines_external_ip:/home/your_google_cloud_user_name

  • You can use this scp command to copy any file on your machine to your virtual machine and vice-versa
  • To copy from your machine to your virtual machine:

    • The format is as follows:

      scp -i (what ssh key to use) (filename) (username@external_ip):/path_where_you'd_like_to_save_your_file

      • In the above case:
        • what ssh key to use = google_cloud_key
        • filename = rqs.txt
        • path_where_you'd_like_to_save_your_file = home/your_google_cloud_user_name
  • To copy from your virtual machine to your machine:

    • The format is as follows:

      scp -i (what ssh key to use) (username@external_ip):/path_where_your_file_is/(filename) path_where_you'd_like_to_save_the_file_on_your_own_machine

• This will now install all needed python packges for machine learning
• Don't worry about any errors/warnings, as along as it the end of the output the screen lists out all the packages that have been installed

Connect to your virtual machine

Start up your virtual environment, source clouds/bin/activate

Run the following command: jupyter notebook --no-browser --port=8887

Now switch back to your own machine and run the following command:

ssh -i ~/.ssh/google_cloud_key -N -L localhost:8888:localhost:8887 your_google_cloud_user_name@your_machines_external_ip

Now open your browser and type in localhost:8888

It will ask you for a token, go back to your virtual machine screen and find the following line on the screen http://localhost:8887/?token=*************************************

Copy everything after the = sign and paste that into the text box on the screen asking you for a token

• When you want to connect again, go back to the dashboard, select your virtual machine and hit start

• Find the Disks button on the left hand side and click it
• Select the disk associated with your machine and click the three dots button under Actions
• Select "Create Snapshot" and name your snapshot something that will allow you to remember when you created this snapshot
• Once the snapshot has finished, delete your instance
• To re-create your instance, find the VM instances button and click on it
• Click the "new" button
• Configure the machine how'd you like it to be configured
  • The standard configuration is:
    • Machine Type: n1-highmem-2 (2 vCPUs, 13 GB memory)
    • CPU platform: Intel Haswell
    • GPUs: 1 x NVIDIA Tesla K80
      • To add the GPU you have to hit the CPU platform and GPU button and then hit the Add GPU button
    • Whatever your appropriate zone is
• Under the Boot Disk option you hit the Change button
• Hit the Snapshots button in the new menu and select the appropriate snapshot you created
• Finally you hit the Create button