Compiling Python files to an exe
Python 3.6 Note
Please note that as of 2017-03-31 this process will not work if the machine is running Python 3.6.
Why might I want to do this?
Note: this is optional to model deployment. When working in with production environments, occasionally there are machines that one has to keep relatively pure. In other words, perhaps you don't want to install a Python (or more specifically, a Python interpreter) on that machine.
But what if you want to run a Python-based predictive model on that same machine, and tie it into nightly ETL processes? That's why you'd want to compile your .py file (which runs your model, for example) to an exe--if you're using Windows--such that it's the only thing you place on the machine that you want to keep pure. If you're using a OS X, one can compile to an .app file.
How does this relate to healthcare.ai?
After you've been able to deploy predictions from your model back to SQL Server, you may want to compile your .py file to an exe.
The workflow of saving and compiling a model to exe
1) Train and save your model by running your
deploy .py file with the
use_saved_model=False argument, such that two pkl files are
2) checked that you can run the model from from the pkl files, by
use_saved_model=True and running your .py script again
3) Note that you should see more rows pushed to SQL Server for 1) and
2). If you didn't see new rows, something is wrong--fix your
deploy.py file before proceeding.
4) If you did see new rows inserted into the database, leave
use_saved_model=True in your script and do the following in
- Install pyinstaller via `conda install -c acellera pyinstaller=3.2.3` - Then run
pyinstaller.exe --noconfirm --log-level=WARN --clean --nowindow --hidden-import=sklearn.tree._utils --name name_of_executable "C:\Users\name\deploy_script.py" --distpath "C:\Users\name\compiled_model_folder" # Copy pkl files (that represent the model) into exe directory just created xcopy "C:\Users\name\working_directory"*.pkl "C:\Users\name\compiled_model_folder" /Y