AWK: convert into lower or upper cases

In order to convert a bash variable to lower case with awk, just use this command:

echo "$a" | awk '{print tolower($0)}'

If you want to convert the content of a file (called data.csv) to lower case:

awk '{print tolower($0)}' data.csv

Of course to convert into upper case, simply use the function toupper() instead of tolower().

Note also that a better tool to avoid issues with special characters might be the tr unix command:

tr [:upper:] [:lower:] < input
Posted in Linux | Tagged , | Leave a comment

How to sort a dictionary by values in Python

By definition, dictionary are not sorted (to speed up access). Let us consider the following dictionary, which stores the age of several persons as values:

d = {"Pierre": 42, "Anne": 33, "Zoe": 24}

If you want to sort this dictionary by values (i.e., the age), you must use another data structure such as a list, or an ordered dictionary.

Use the sorted function and operator module

import operator
sorted_d = sorted(d.items(), key=operator.itemgetter(1))

Sorted_d is a list of tuples sorted by the second element in each tuple. Each tuple contains the key and value for each item found in the dictionary. If you look at the content of this variable, you should see:

[ ('Zoe', 24), ('Anne', 33), ('Pierre', 42)]

Use the sorted function and lambda function

If you do not want to use the operator module, you can use a lambda function:

sorted_d = sorted(d.items(), key=lambda x: x[1])
# equivalent version
# sorted_d = sorted(d.items(), key=lambda (k,v): v)

The computation time is of the same order of magnitude as with the operator module. Would be interesting to test on large dictionaries.

Use the sorted function and return an ordered dictionary

In the previous methods, the returned objects are list of tuples. So we do not have a dictionary anymore. You can use an OrderedDict if you prefer:

>>> from collections import OrderedDict
>>> dd = OrderedDict(sorted(d.items(), key=lambda x: x[1]))
>>> print(dd)
OrderedDict([('Pierre', 24), ('Anne', 33), ('Zoe', 42)])

Use sorted function and list comprehension

Another method consists in using list comprehension and use the sorted function on the tuples made of (value, key).

sorted_d = sorted((value, key) for (key,value) in d.items())

Here the output is a list of tuples where each tuple contains the value and then the key:

[(24, 'Pierre'), (33, 'Anne'), (42, 'Zoe')]

A note about Python 3.6 native sorting

In previous version on this post, I wrote that “In Python 3.6, the iteration through a dictionary is sorted.”. This is wrong. What I meant is that in Python 3.6 dictionary keeps insertion sorted.

It means that if you insert your items already sorted, the new python 3.6 implementation will be this information. Therefore, there is no need to sort the items anymore. Of course, if you insert the items randomly, you will still need to use one of the method mentioned above.

For instance, taking care of the age, we now create our list as follows (sorting by ascending age):

d = {("Zoe": 24)}
d.update({'Anne': 33})
d.update({'Pierre': 42})

Now you can iterate through the items and they will be in the same order as in the creation of the dictionary. So you can just create a list from your items very easily:

Out[15]: [('Zoe', 24), ('Anne', 33), ('Pierre', 42)]


Here is a quick benchmark made using the small dictionary from the above examples. Would be interesting to redo the test with a large dictionary.

What you can see is that the native Python dictionary sorting is pretty cool followed by the combination of the lambda + list comprehension method. Overall using one of these methods would be equivalent though (factor 2/3 at most).

This image was created with the following code.

import operator                                                  
import pylab
from easydev import Timer
times1, times2, times3 = [], [], []
d = {"Pierre": 42, "Anne": 33, "Zoe": 24}
for j in range(20):
    N = 1000000
    with Timer(times3):
        for i in range(N):
         sorted_d = sorted((key, value) for (key,value) in d.items())
    with Timer(times2):
        for i in range(N):
            sorted_d = sorted(d.items(), key=lambda x: x[1])
    with Timer(times1):
        for i in range(N):
            sorted_d = sorted(d.items(), key=operator.itemgetter(1))
pylab.boxplot([times1, times2, times3])
pylab.xticks([1,2,3], ["operator", "lambda", "list comprehension and lambda"])
pylab.ylabel("Time (seconds) 1 million sorting \n (repeated 20 times)")
pylab.title("Performance sorted dictionary by values")
Posted in Python, Uncategorized | Tagged , | 6 Comments

Python: how to copy a list

To explain how to create a copy of a list, let us first create a list. We will use a simple list of 4 items:

list1 = [1, 2, "a", "b"]

Why do we want to create a copy anyway ? Well, because in Python, this assignement creates a reference (not a new independent variable):

list2 = list1

To convince yourself, change the first item of list2 and then check the content of list1, you should see that the two lists have been modified and contain the same items.

So, to actually copy a list, you have several possibilities. From the simplest to the most complex:

  • you can slice the list.
    list2 = 1ist1[:]
  • you can use the list() built in function
    list2 = list(1ist1)
  • you can use the copy() function from the copy module. This is slower than the previous methods though.
    import copy
    list2 = copy.copy(list1)
  • finally, if items of the list are objects themselves, you should use a deep copy (see example below):
    import copy
    list2 = copy.deepcopy(list1)
  • To convince yourself about the interest of the latter method, consider this list:

    import copy
    list1 = [1, 2, [3, 4]]
    list2 = copy.copy(list1)
    list2[2][1] = 40

    you should see that changing list2, you also changed list1. If this is not the intended behviour, you should consider using the deepcopy.

    Posted in Python | Tagged , | 2 Comments

    Python: ternary operator

    In C language (and many other languages), there is a compact ternary conditional operator that is a compact if-else conditional construct. For instance, in C, a traditional if-else construct looks like:

    if (a &gt; b) {
        result = x;
    } else {
        result = y;

    and the equivalent ternary operator looks like:

    result = a>b ? x : y;

    As in the if-else code, only one expression x or y is evaluated.

    In Python, from version 2.5, you would write:

    results = x if a > b else y

    More formally the ternary operator is written as:

    x if condition else y

    So condition is evaluated first then either x or y is returned based on the boolean value of condition.

    You can use ternary operator within list comprehension. For example:

    [1 if item > else -1 for item in [0,1,-5,2]]
    Posted in Python | Leave a comment

    Difference between __repr__ and __str__ in Python

    When implementing a class in Python, you usually implement the __repr__ and __str__ methods.

    1. __str__ should print a readable message
    2. __repr__ should print a message that is unambigous (e.g. name of an identifier, class name, etc).

    You can see __str__ as a method for users and __repr__ as a method for developers.

    Here is an implementation example for a class that simply stores an attribute (data).

    class Length():
        def __init__(self, data):
   = data

    __str__ is called when a user calls the print() function while __repr__ is called when a user just type the name of the instance:

    >>> l = Length([1,2,3])
    >>> print(l)    # should call __str__ if it exists
    <__main__.Length at 0x7faf240acc18>
    >>> l
    <__main__.Length object at 0x7faf240acc18>

    By default when no __str__ or __repr__ methods are defined, the __repr__ returns the name of the class (Length) and __str__ calls __repr__.

    Now, let us define the __repr__ method ourself to be more explicit:

    class Length():
        def __init__(self, data):
   = data
        def __repr__(self):
            return "Length(%s) " % (len(

    we could use it as follows:

    >>> l = Length([1,2,3])
    >>> print(l)     # calls __str__
    >>> l            # calls __repr__
    Length(3, 140175447410224)

    When using the print() function in Python, the __str__ is called (if found) and otherwise, __repr__.

    class Length():
        def __init__(self, data):
   = data
        def __repr__(self):
            return "Length(%s, %s) " % (len(, id(self))
        def __str__(self):
            return "Length(%s) " % (len(

    so now __repr__ and __str__ have different behaviours:

    >>> l = Length([1,2,3])
    >>> print(l)     # calls __str__
    >>> l            # calls __repr__
    Length(3, 140175447410224)
    Posted in Python | Tagged , | 3 Comments

    python: how to merge two dictionaries

    Let us suppose two dictionaries storing ages of different individuals:

    list1 = {"Pierre": 28, "Jeanne": 27}
    list2 = {"Marc": 32, "Helene": 34}

    If you do mind losing the contents of either list1 or list2 variable, you can update one of the other as follows:


    Now list1 variable contains:

    {"Pierre": 28, "Jeanne": 27, "Marc": 32, "Helene": 34}

    while list2 is unchanged.

    Usually, this is not what you want though. Instead, you would prefer to create a third variable keeping list1 and list2 unchanged.

    In Python 3.5 or greater, you can use the following syntax:

    fulllist = {**list1, **list2}

    In Python 2 or 3.4 and below, you need to copy one of the variable and update it:

    full_list = list1.copy()  # this keeps list1 unchanged
    full_list.update(list2)   # inplace update of the variable full_list

    The second method is more generic and would be more backward compatible (if you plan to provide your code to Python 2 users. Indeed, it would work for Python 2 and 3. However, it would be slower for Python 3.5 users (and above).

    Posted in Python | Tagged | Leave a comment

    Search for a pattern in a set of files using find and grep commands

    A common task for developers is to search for a pattern amongst a bunch of files that are in different directories.

    For instance, you are looking for the pattern “import sys” within a set of Python files. Those files are in sub directories mixed with other documents.

    You can use the find command (to look for files ending with the py extension) and redirect those files to the grep command to search for the pattern “import sys” within all files found by the find command:

    find . -name "*py" | xargs grep "import sys"

    Note the double quotes and the use of the xargs command to scan the content of the files (not their names).

    Of course, you can use all kind of wildcards:

    find . -name "*py" | xargs grep "import sys"

    Posted in Linux | Tagged , | Leave a comment

    okular: export annotations in the PDF file

    One open source software to add annotations under Linux is okular (

    One can add annotations easily (go to Tools, tick review, or just type F6).

    Then, it is time to save your document or to send it to a collaborator but wait a minute: we do not see the annotations when using acroread reader !! No worries, plenty of resources tell you to go to File/save as

    Seems to work indeed. You quit, open the file and there you can see the annotations. Now, I use xpdf to read the PDF file. And here nothing. Oh, and I send the PDF to a journal review; they include the PDF inside another one and there no annotations either….

    Final solution: instead of File/Save as, just print the document in … a PDF file: File/Print

    This worked for me.

    Posted in Linux | Tagged | 3 Comments

    No more space left on /tmp under Fedora

    Under Fedora, one of my software requires more than 4Gb of temporary space and I realised that the /tmp directory is limited to 4Gb. In order to increase the /tmp directory, just edit the /etc/fstab file and add this line (to extend to 8Gb instead of 4Gb):

     none /tmp tmpfs size=8G 0 0

    Then, as root:

    mount -a

    That’s it. You can check that you have now 8Gb available in /tmp by typing

    df -h
    Posted in Linux | Leave a comment

    blasr (pacbio) installation on fedora box

    I wanted to use blasr tool for Pacbio mapping and had difficulties in using or installing the tool. I first use a local installation of the tool on the provided cluster and it look like the installation was quite old. I then use a bioconda version. The latest version (5.3) was not working on my system (missing library) and this was reported. The previous version worked (5.2) but was missing a needed configuration flag. So, I decided to follow the instructions from pacbio to install a local version. That was not straightforward but finally got it to work. My platform is Fedora 23 and the instructions were given for ubuntu or centos 6.

    First, download the source code:

    git clone
    cd pitchfork
    make PREFIX=/tmp/mybuild blasr

    This command installs zlib, bzip2, boost and hdf5 to start with.
    The first issue arised from an error in the compilation of the hdf5 dependency due to missing iostream in the compilation of the hdf5 library

    H5 Attribute.cpp fatal error iostream.h no such file

    The solution was to edit the Makefile in workspace/hdf5-1.8.16/c++
    and to comment this line (adding a # character in front))


    Come back to the main pitchfork directory, type the make command again (see above). Another failure due to similar compilation error related to the namespace occured. Again, I edited the Makefile in
    workspace/hdf5-1.8.16/c++/test/Makefile and commented the same line.

    Next, I got a linking issue

    /usr/bin/ld: cannot find -lstd++

    This was solved by installing the stdc++ static library. I figure out the solution by typing:

    ld -lstdc++

    to see that none of the standard path could find the library despite the presence of the /usr/lib/

    yum install libstdc++-static

    Again, type

    make hdf5 PREFIX=/tmp/mybuild

    I got

    ptableTest.cpp:20:19: error: expected namespace-name before ‘;’ token
     using namespace H5;
    Makefile:745: recipe for target 'ptableTest.o' failed
    make[5]: *** [ptableTest.o] Error 1

    so again, needed to find the culprit: a Makefile and it was in
    In workspace/hdf5-1.8.16/c++/hl/test/Makefile . Again comment the same line as shown above.

    Back to blasr, the make then failed when trying to install ncurses.

    Here I tried a different strategy and tried to use the packages installed with my conda environement. To do so, I edited the and added:


    Then, same issue with samtools, so added


    Then, there was an error in the ./bin/pitchfork module due to a Python 3 issue (I had already switch all print “” to print(“”). Here, the issue was

    ptableTest.cpp:20:19: error: expected namespace-name before ‘;token
     using namespace H5;
    Makefile:745: recipe for target 'ptableTest.o' failed
    make[5]: *** [ptableTest.o] Error 1

    Just replace _out[0] with str(_out[0])

    Then, blasr compiles successfully…time to run it; The executable seems to be in workspace/blasr/blasr:

    ./workspace/blasr/blasr: error while loading shared libraries: cannot open shared object file: No such file or directory

    Here, you need to add a bunch of path to your LD_LIBRARY_PATH:

    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PF/blasr_libcpp/hdf/
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PF/blasr_libcpp/alignment
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PF//blasr_libcpp/pbdata
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PF/pbbam/build/lib/
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$PF/hdf5-1.8.16/c++/src/.libs/
    Posted in biology | Tagged , | Leave a comment