ML4T Software Setup
Most of the projects in this class will be graded automatically. As of the summer 2017 semester, we are providing the grading scripts with the template code for each of the projects, so that students can test their code to make sure they are API compatible. Georgia Tech also provides access to four servers that have been configured to be identical to the grading environment, specifically in terms of operating system and library versions. Since these servers have already been configured with all necessary libraries, setup has been reduced to simply checking out a single git repository, which will be covered below. For students with sporadic internet access who would like a local installation of the software, including libraries with the correct versions, the instructions from previous semesters are available here: ML4T_Software_Installation.
- Your code MUST run properly on the Georgia Tech provided servers, and your code must be submitted to T-square. If you do not test your code on the provided machines it may not run correctly when we test it. If your code fails to run on the provided servers, you will not get credit for the assignment. So it is very important that you ensure that you have access to, and that your code runs correctly on, these machines. If you would like to develop on your personal machine and are comfortable installing libraries by hand, here are the specific versions of the libraries we use (Please note that these are not the most recent versions for many of these libraries):
- Python 2.7.6
- numpy 1.10.2
- matplotlib 1.5.0
- pandas 0.17.1
- scipy 0.16.1
- dateutil 2.4.2
- pytest 2.5.1
- We use a specific, static dataset for this course, which is provided as part of the repository detailed below. If you download your own data from Yahoo (or elsewhere), you will get wrong answers on assignments.
- We reserve the right to modify the grading script while maintaining API compatibility with what is described on the project pages. This includes modifying or withholding test cases, changing point values to match the given rubric, and changing timeout limits to accommodate grading deadlines. The scripts are provided as a convenience to help students avoid common pitfalls or mistakes, and are intended to be used as a sanity check. Passing all tests does not guarantee full credit on the assignment, and should be considered a necessary but not sufficient condition for completing an assignment.
Access to machines at Georgia Tech
There are 4 machines that will be accessible to students enrolled in the ML4T class via ssh. These machines may not be available until the second week of class; we will make an announcement once they are ready, and if at that time you are still unable to log in, please contact us. If you are using a Unix based operating system, such as Ubuntu or Mac OS X, you already have an ssh client, and you can connect to one of the servers by opening up a terminal and typing:
xhost + ssh -X gtname@buffet0X.cc.gatech.edu
replacing the X in buffet0X with 1-4, as detailed below. You will then be asked for your password and be logged in. Windows users may have to install an ssh client such as putty. In order to distribute workload across the machines, please use the specific machines as follows:
- buffet01.cc.gatech.edu if your last name begins with A-G
- buffet02.cc.gatech.edu if your last name begins with H-N
- buffet03.cc.gatech.edu if your last name begins with O-U
- buffet04.cc.gatech.edu if your last name begins with V-Z
These machines use your GT login credentials.
NOTE: We reserve the right to limit login access or terminate processes to avoid resource contention during grading, although we will endeavor to limit such interruptions.
Getting code templates
After you've successfully logged in, you will need to clone the following git repository containing all of the template code and data into your home directory: https://github.gatech.edu/bhrolenok3/ML4T_2017Summer. You can do this with the following command:
again providing your GT login credentials when asked for. For the remainder of these instructions, we'll assume you checked out the repository into your home directory, and that you did not change the name of the folder.
Running the grading scripts
The repository you've just cloned contains the grading scripts, data, and template code for all assignments. To complete the assignments you'll need to modify the templates according to the assignment description. You can do this on the buffet0X machines directly using a text editor such as gedit, nano, or vim. Or you can copy the file to your local machine, edit them in your favorite text editor or IDE, and upload them back to the server. Make sure to test run your code on the server after making changes to catch any typos or other bugs.
To test your code, you'll need to set up your PYTHONPATH to include the grading module and the utility module util.py, which are both one directory up from the project directories. Here's an example of how to run the grading script for the first assignment:
PYTHONPATH=../:. python grade_analysis.py
which assumes you're typing from the folder ML4T_2017Summer/mc1p1_assess_portfolio/. This will print out a lot of information, and will also produce two text files: points.txt and comments.txt, which summarize the output, including any errors or failed test cases.
Updating the repository
We will periodically update the repository throughout the semester. When this happens, we will make a note of it on the Repository Update Page, which you should check regularly. Below are instructions for updating your copy of the repository.
Note: these instructions are for students who have not committed to their repository, added a different origin, or any other advanced git techniques. If you have done this, some quick googling should resolve any questions you have.
From here on, we'll assume you've checked out the repository, and may have made some modifications you'd like to keep. First things first, figure out what you have changed since you originally pulled the repo. From the your code root directory (e.g., ML4T_2017Summer/), run the following command:
Look at the list have files that have changed and make sure it makes sense. For example, if you've only modified the python file for the first assignment, the output may look something like this:
bhrolenok3@buffet02:~/ML4T_2017Summer$ git status On branch master Your branch is up-to-date with 'origin/master'. Changes not staged for commit: (use "git add <file>..." to update what will be committed) (use "git checkout -- <file>..." to discard changes in working directory) modified: mc1p1_assess_portfolio/analysis.py no changes added to commit (use "git add" and/or "git commit -a")
You may see a few lines after this under the heading "Untracked files", these are safe to ignore. They are just files that aren't part of the repository (temporary backups, .pyc files, notes, etc). If you see any modified files that you don't remember editing, you can look at the exact differences by using the following git command:
git diff <filename>
replacing <filename> with the name of the file that's been marked as modified. Following the example earlier, here's what running that command looks like for the analysis.py changes I made:
bhrolenok3@buffet02:~/ML4T_2017Summer$ git diff mc1p1_assess_portfolio/analysis.py diff --git a/mc1p1_assess_portfolio/analysis.py b/mc1p1_assess_portfolio/analysis.py index 9a9c1c6..10d422e 100644 --- a/mc1p1_assess_portfolio/analysis.py +++ b/mc1p1_assess_portfolio/analysis.py @@ -24,7 +24,7 @@ def assess_portfolio(sd = dt.datetime(2008,1,1), ed = dt.datetime(2009,1,1), \ port_val = prices_SPY # add code here to compute daily portfolio values # Get portfolio statistics (note: std_daily_ret = volatility) - cr, adr, sddr, sr = [0.25, 0.001, 0.0005, 2.1] # add code here to compute stats + cr, adr, sddr, sr = [0.50, 0.002, 0.0010, 4.2] # twice as good! # Compare daily portfolio value with SPY using a normalized plot if gen_plot:
lines with - have been removed, lines with + have been added, so this output means I changed one line in the file, changing the
cr, adr, sddr, sr variables. You'll be able to scroll up and down through the changes using your arrow keys, and you'll need to hit the q key to get back to the command line. Once you've identified all the changed files, use scp (or WinSCP or the ssh client of your choice) to copy the files you'd like to keep to your local computer. Now, you can stash all the changes you've made on your copy of the repo on buffet0x using
git stash, which, following our example, will look something like this:
bhrolenok3@buffet02:~/ML4T_2017Summer$ git stash Saved working directory and index state WIP on master: a97a488 Grading script for mc1p1, initial commit HEAD is now at a97a488 Grading script for mc1p1, initial commit
Now you can safely pull down all the changes that have been made to the repo since the last time. Do that using
bhrolenok3@buffet02:~/ML4T_2017Summer$ git pull Enter passphrase for key '/home/bhrolenok3/.ssh/id_rsa': DISPLAY "(null)" invalid; disabling X11 forwarding remote: Counting objects: 7, done. remote: Compressing objects: 100% (7/7), done. remote: Total 7 (delta 0), reused 0 (delta 0), pack-reused 0 Unpacking objects: 100% (7/7), done. From github.gatech.edu:tb34/ML4T_2017Summer 228f9ec..803f0be master -> origin/master Updating 228f9ec..803f0be Fast-forward mc3p1_assess_learners/Data/winequality-red.csv | 1599 ++++++++++++ mc3p1_assess_learners/Data/winequality-white.csv | 4898 +++++++++++++++++++++++++++++++++++++ mc3p1_assess_learners/Data/winequality.names.txt | 72 + 3 files changed, 6569 insertions(+) create mode 100644 mc3p1_assess_learners/Data/winequality-red.csv create mode 100644 mc3p1_assess_learners/Data/winequality-white.csv create mode 100644 mc3p1_assess_learners/Data/winequality.names.txt
This should be similar for everyone, since the only time the remote repository is updated is when we (TAs/Professor Balch) make changes. At this point, you'll have all the new changes to the repository. From here you can 1) start working from scratch on the current assignment (safest option), 2) copy back the modified files using scp (verify by hand), or 3) use
git stash to apply the changes to the new repository. Option 2 should be safe and quick in most instances, and if you're not comfortable with
git and the command line may be the easiest. You'll have to check the differences between any files you overwrite when you copy them back to buffet0x, which you can do easily with the
git diff command described earlier. Option 3 handles all of these things using
git's own tools. To apply your stashed changes from earlier, you can simply call
git stash pop:
bhrolenok3@buffet02:~/ML4T_2017Fall$ git stash pop On branch master Your branch is up-to-date with 'origin/master'. Changes not staged for commit: (use "git add <file>..." to update what will be committed) (use "git checkout -- <file>..." to discard changes in working directory) modified: mc1p1_assess_portfolio/analysis.py
which tells you the status of the repo after applying all your changes, which you should double check makes sense using
git diff as before. If you see any "conflicts" or error messages when applying your stashed changes, you'll need to go back over them by hand. Since you have a backup of your files, you can always wipe out the repo and start from a clean slate.