How can I use both Anaconda versions (2.7 & 3.5)?

Question

I was using the Anaconda 3.5 distro in a Windows 10 machine. Due to dependencies in libraries that I want to work with, I had to have the 2.7 version installed as well.

The good news is that the libraries I needed can now work with the 2.7 version smoothly and Visual Studio 2015 automagically detected my new Python environment.

The problem comes when using the command line. Upon issuing the command

conda info --envs

I get

root                  *  C:\Users\norah\Anaconda2

i.e. a single environment (to my understanding and search so far, according to this I should see two envs listed). This means I can't use conda to acquire new packages for my Python3.5 installation, at least not at the command line as I used to since conda only refers to Python2.7. The same goes for the GUI version, Anaconda navigator (I'm not very fond of the GUI version but I tried it out).

There's also no way of launching python3 from the command line since

$python

always fires up python2.7 and issuing $python3 or $python3.5 in the command line doesn't seem to work (nor would adding the path of python3 to the system since the actual executable has the same name as python2 i.e. python.exe)

Is my system taken over by Python2.7? Is anyone here using them both smoothly and if so could you please elaborate on how to do that? Was it a "no no" move to install both versions of Anaconda?

Solution 1

You can make Python 3.5 environment with your Anaconda 2.7:

conda create -n py35 python=3.5

Now, activate it:

activate py35

Finally you can install the desired packages:

conda install numpy

or, the whole anaconda:

conda install anaconda

The advantage of this approach is that you can also create Python 3.4 or 3.6 environments. Furthermore, you can create environments with different combinations and versions of libraries.

Actually, it makes sense to create a new environment for each larger project.

Solution 2

I also despise the virtual environment switch that Anaconda tries to force on us. I prefer to have both executables always instantly available from the command line. I'm pretty sure I had this working on a Windows machine once:

1. Install Anaconda2 and Anaconda3 to the C:\ drive as "C:\Anaconda2\" and "C:\Anaconda3\" respectively.
2. Edit your "Path" environment variable (Control Panel -> System and Security -> System -> Advanced system settings -> Environment Variables) and make sure that "C:\Anaconda2;C:\Anaconda2\Scripts;C:\Anaconda2\Library\bin" is in front of "C:\Anaconda3;C:\Anaconda3\Scripts;C:\Anaconda3\Library\bin".
3. Copy and rename the file "C:\Anaconda3\python.exe" to "C:\Anaconda3\python3.exe".
4. Copy and rename the file "C:\Anaconda3\Scripts\conda.exe" to "C:\Anaconda3\Scripts\conda3.exe"
5. Copy and rename any other scripts you might use in "C:\Anaconda3\Scripts\", such as "pip.exe" to "pip3.exe", etc.

Now, when you type "python" or "conda" at the command line you will get the python2 version, and when you type "python3" or "conda3", etc. at the command line you will get the python3 version.

Solution 3

The best way to use both Python 2.7x and Python 3.5x together is Jupyter Notebook.

http://jupyter.org/

You will be able to work on your browser with IPython Notebook style interface which is great for working with scripting languages.

I found some of these videos on YouTube very informative:

1) https://www.youtube.com/watch?v=HW29067qVWk

2) https://www.youtube.com/watch?v=e9cSF3eVQv0

Besides Python 2.7 and 3.5, there are a bunch of other languages that you will be able run with your Jupyter Notebook. The various Kernels are available in this link below:

https://github.com/jupyter/jupyter/wiki/Jupyter-kernels

After installation, as you start your Jupyter Notebook, your browser will open up a new page showing your Home Directory. Among the 4 tabs (Files|Running|Clusters|Conda):

1) The Files tab shows all the files in your Home Directory.

2) Running tab shows all the Notebooks that are running.

3) Clusters tab is provided by IPython parallel.

4) Conda tab is where you need to add the Python version 3.5 (and other languages if needed) to your Jupyter Notebook (If Python 2.7 is default).

If you are interested to try C++ with your Jupyter Notebook, there are a couple of Kernels available.

1) JupyRoot - https://github.com/root-mirror/root/tree/master/bindings/pyroot/JupyROOT

2) Cling - https://github.com/root-mirror/cling

Comments

[Lorah Attkins]

This was my initial approach but issuing the first command you mention lead to "recursion error" :( hence I went with a fresh install

[pylang]

@LorahAttkins This is the best way to manage different versions of python with Anaconda on Windows. If you get an error running the first command, there is an issue with the python installation. Make sure by "clean installation", you uninstalled all other anaconda distributions. Avoid having Anaconda 2.x and Anaconda 3.x installed together; pick one and use envs as shown here.

[Lorah Attkins]

@pylang then it's not OK to run both installers so I should delete one distribution and start all over again. I have to finish a project and I'm afraid messing with the installation right now but will try again in a week or so. +1 here and I'll wait to see if there are any more answers before accepting

[Lorah Attkins]

this looks like something the installers could have done for us right?

[u55]

@LorahAttkins Yes, the Anaconda packagers could have included all the python versions as individually-named executables (or soft links to them), such as python, python2, python2.7, python3, python3.5, etc., as is commonly done by the packagers for most Linux distributions, and possibly even let the user decide which version "python" points to, but it seems the Anaconda devs are too infatuated by Virtualenv.

[pylang]

@LorahAttkins Yes. I recommend, when and if you are able, truly start from scratch and delete all Anaconda distributions. Then install only one. One distribution is sufficient as you can install separate environments. This suggestion is the appropriate method, and I also recommend looking at the SO docs on this topic for assurance stackoverflow.com/documentation/python/868/virtual-environme‌​nts/…. Consider using a Python 3 distribution with Py 2 env since support will end eventually pythonclock.org.

[pylang]

@u55 I am curious why are you opposed to Anaconda's virtual env style of handling python versions?

[u55]

@pylang Virtualenv is great for testing code against every python version under the sun, but I don't want to have to use it for daily work because activate py35 && python and deactivate is too much typing for me. --Not joking. I much prefer to be able to use python or python3 whenever needed. Also, having both executables simultaneously makes it much easier to set up build commands for text editors/IDE's, and makes it much easier to call one version of python from a different version of python using "subprocess".

[pylang]

I think I can commiserate with your second argument. Thanks.

[jxn]

Unfortunately, this approach means that one will have to reinstall the required packages for python3 despite already having anaconda?

[Mike Müller]

You always need to install a package for each Python version no matter approach you use. Python 2 packages of NumPy, SciPy etc. will never work in Python 3.

[rainman]

While using the command conda install anaconda, is it possible to specify the version of anaconda?

[Mike Müller]

Just specify the version number conda install anaconda=5.3.0.

[qqqqq]

@u55 using this approach, during installation of anaconda when prompted to select (check boxes)between modifying PATH and Register Anaconda( the default). Should I turn off both? Note that I have already version 2 installed and it is on my PATH. I am asking it in context of second installation (version 3 )?

[u55]

@qqqqq Sorry for the late reply. It doesn't matter if you let the installation prepend to the Path environment variable or not since you are editing it by hand in step 2. As for "registering" a default python version, the installation instructions say that this is used by Visual Studio PyCharm, Wing IDE, PyDev, etc. If you don't use these Windows-centric programs then you'll never care about the Windows registry keys.