Is there a simple way to install Apache Mahout on Windows or Mac without the need of Hadoop? I heard there is a library called Taste which Mahout is based on... but I did not find it on the web
I believe you won't be able to decouple mahout and hadoop. Mahout uses the hadoop API in the code and it also ships with the hadoop jar files. And this is irrespective of whether you run mahout on Windows or Linux
Mahout has a lot of code. Some requires Hadoop to run, some doesn't. If you don't need the bits that use Hadoop, you don't need Hadoop.
And yes in particular, some of the collaborative filtering code came from "Taste" (I'm the author) which is not distributed, not Hadoop-based. (There are also Hadoop-based recommenders inside Mahout.)
The answer is you don't have to do anything with Hadoop, therefore, to "install" Mahout by itself, if you're not using these bits. The Maven build script will download the Hadoop libraries for you just for compilation purposes.
It is true that you will need a hadoop jar in your system to compile Mahout, but you don't need to actually run Hadoop for big chunks of Mahout. That jar is downloaded automagically so you don't have to do anything explicit to get it.
I work on Mahout on my Mac all the item without running Hadoop. In fact, with maven doing the dependency work for me, I don't even really have any awareness of the dependency impinging on my work which is almost all lately on the non-Hadoop code.
It should be about as easy on Windows, but you almost certainly will need cygwin and you will run into problems if only because Windows is, well, so much like Windows. On a Mac or particularly on Linux, you won't have any problems of that sort at all.