The advantage to the Mat. packages — you can compile, say, the same will dotnet, and it would not require each user to buy and put Matlab.
\r
Can be programmed in parallel, and cluster.
\r
Data edit is possible, but medlennovarki — I eventually settled on Scala, there all buns of modern languages (for example, I love the tuples), and speed is almost seeplusplus.
Python commercials 40 times slower. In fact, as Matlab, as I remember it. This significantly limits the range to conveniently solve the language tasks.
For beautiful output and heaps of built-in scientific primitives the same GnuScienceLibrary (vkluchaet gnuplot as its component) prebendary to all possible languages, as far as I know...
\r
In General, I use a dynamic language (although Ruby and not Python) to fast (no recompiles) debugging algorithm on a small dataset, and then write the final version in Scala and compile to normal .jar.