Quick Anaconda Review (for Udacity Deep Learning ND Foundations Program)

Anaconda Review

# Anaconda

 * Distribution of libraries and software specifically built for data science.
 * Conda = package and environment manager 
 * Best practices = create a new enviornment for each chapter
 * Then use "conda install" to install dependencies (modules)
 * Use "conda list" to view all installed package in the environment
 * You can also export the environment (package listing) to a file. Share that with others to enable them to replicate your environment when running code.

### Installing Anaconda

1. Available for Windows/MacOS/Linux. Installers here: [https://www.continuum.io/downloads](https://www.continuum.io/downloads)
2. Complements existing Python dist on computer. _Default Python used by programs will be that installed by Anaconda_
3. Udacity recommends: Python 3.5, 64-bit installer  
4. I have: Anaconda 4.3.9 for OSX (Python 2.7, 64-bit) 
    - I used ```conda update anaconda``` to update it from my previous 4.1.x version but it had conflicts with updating python versions
    - Instead I had to delete/move the "anaconda" directory from my home and then reinstall a fresh version from the downloaded package 
5. Now have: Anaconda 4.3.x on Python 3.6 _however this removes previously created environments so I need to redo those_


### Managing Packages

Once you have Anaconda installed, managing packages is fairly straightforward.

1. Install package: e.g.,  ```conda install numpy``` 
2. Install multiple packages: e.g., ```conda install numpy scipy pandas``` 
3. Install specific version of package e.g., ```conda install numpy=1.10```
4. Install dependencies e.g., ```conda install scipy``` also installs _numpy_ which _scipy_ depends on
5. Update package: ```conda update tensorflow``` (for one) or ```conda update --all``` to update all packages in that environment
6. List installed packages in environment: ```conda list``` 
7. Remove package: ```conda remove tensorflow```
8. Search for package ```conda search beautifulsoup```


### Managing Environments

1. Create a new "named" environment:  ```conda create -n my_new_env numpy tensorflow``` (where -n provides arg for name)
2. Specify default python install for environment during creation: ```conda create -n my_new_env python=3 numpy tensorflow``` (use version can be general e.g., "2" or "3" - or specific e.g., "3.3")
3. Activate environment: ```source activate my_new_env``` (environment name will appear in terminal prompt as prefix)
4. Deactivate environment: ```source deactivate my_new_env```
5. Running ```conda install``` (or other package management commands) within an activated environment ensures changes are localized to that environment.

### Saving and Loading Environments

1. Save environment: as metadata in _YAML_ format => ```conda env export > environment.yaml```
2. Load environment: recreate from _YAML_ => ```conda env create -f environment.yaml```
3. Note to self: save env to _yaml_ for any project by default. If you need to upgrade or move Anaconda and lost your previous created environments, this lets you restore them from files.
4. Listing enviroments: ```conda env list```
5. Remove outdated environments ```conda env remove -n env_name```

### Best Practices

1. Using environments: _recommends created pre-defined py2 and py3 named environments corresponding to Python2 and Python3 versions. Simplies the ability to do general purpose coding for each case_
2. Sharing environments: _recommends always creating saved env file for projects to allow others to restore/replicate your work._ (Note to self: let's use convention of adding _conda_env.yaml_ as default named file for project )
3. Conda doc: [website](http://conda.pydata.org/docs/using/index.html)
4. Conda myths: [JVP article](https://jakevdp.github.io/blog/2016/08/25/conda-myths-and-misconceptions/)