How can I parse an arff file without using external libraries in Python

Question

I need to parse an arff file like the following without using any external libraries. I am not sure how I can related the attributes to the numerical values. Like how can I say the first numerical value in each row is the age while the second one is the sex? Can you also link me to some python codes which are for parsing similar scenarios?

@relation cleveland-14-heart-disease
@attribute 'age' real
@attribute 'sex' { female, male}
@attribute 'cp' { typ_angina, asympt, non_anginal, atyp_angina}
@attribute 'trestbps' real
@attribute 'chol' real
@attribute 'fbs' { t, f}
@attribute 'restecg' { left_vent_hyper, normal, st_t_wave_abnormality}
@attribute 'thalach' real
@attribute 'exang' { no, yes}
@attribute 'oldpeak' real
@attribute 'slope' { up, flat, down}
@attribute 'ca' real
@attribute 'thal' { fixed_defect, normal, reversable_defect}
@attribute 'class' { negative, positive}
@data
63,male,typ_angina,145,233,t,left_vent_hyper,150,no,2.3,down,0,fixed_defect,negative
37,male,non_anginal,130,250,f,normal,187,no,3.5,down,0,normal,negative
41,female,atyp_angina,130,204,f,left_vent_hyper,172,no,1.4,up,0,normal,negative
56,male,atyp_angina,120,236,f,normal,178,no,0.8,up,0,normal,negative
57,female,asympt,120,354,f,normal,163,yes,0.6,up,0,normal,negative
57,male,asympt,140,192,f,normal,148,no,0.4,flat,0,fixed_defect,negative
56,female,atyp_angina,140,294,f,left_vent_hyper,153,no,1.3,flat,0,normal,negative
44,male,atyp_angina,120,263,f,normal,173,no,0,up,0,reversable_defect,negative
52,male,non_anginal,172,199,t,normal,162,no,0.5,up,0,reversable_defect,negative

Here's a sample code I have written:

arr=[]
arff_file = open("heart_train.arff")
count=0
for line in arff_file:
        count+=1
        #line=line.strip("\n")
        #line=line.split(',')
        if not (line.startswith("@")):
                if not (line.startswith("%")):
                        line=line.strip("\n")
                        line=line.split(',')
                        arr.append(line)



print(arr[1:30])

However the output is very different than I expect it to be:

[['37', 'male', 'non_anginal', '130', '250', 'f', 'normal', '187', 'no', '3.5', 'down', '0', 'normal', 'negative'], ['41', 'female', 'atyp_angina', '130', '204', 'f', 'left_vent_hyper', '172', 'no', '1.4', 'up', '0', 'normal', 'negative'], ['56', 'male', 'atyp_angina', '120', '236', 'f', 'normal', '178', 'no', '0.8', 'up', '0', 'normal', 'negative'], ['57', 'female', 'asympt', '120', '354', 'f', 'normal', '163', 'yes', '0.6', 'up', '0', 'normal', 'negative'], ['57', 'male', 'asympt', '140', '192', 'f', 'normal', '148', 'no', '0.4', 'flat', '0', 'fixed_defect', 'negative'], ['56', 'female', 'atyp_angina', '140', '294', 'f', 'left_vent_hyper', '153', 'no', '1.3', 'flat', '0', 'normal', 'negative'], ['44', 'male', 'atyp_angina', '120', '263', 'f', 'normal', '173', 'no', '0', 'up', '0', 'reversable_defect', 'negative'], ['52', 'male', 'non_anginal', '172', '199', 't', 'normal', '162', 'no', '0.5', 'up', '0', 'reversable_defect', 'negative'], ['57', 'male', 'non_anginal', '150', '168', 'f', 'normal', '174', 'no', '1.6', 'up', '0', 'normal', 'negative'], ['54', 'male', 'asympt', '140', '239', 'f', 'normal', '160', 'no', '1.2', 'up', '0', 'normal', 'negative'], ['48', 'female', 'non_anginal', '130', '275', 'f', 'normal', '139', 'no', '0.2', 'up', '0', 'normal', 'negative'], ['49', 'male', 'atyp_angina', '130', '266', 'f', 'normal', '171', 'no', '0.6', 'up', '0', 'normal', 'negative'], ['64', 'male', 'typ_angina', '110', '211', 'f', 'left_vent_hyper', '144', 'yes', '1.8', 'flat', '0', 'normal', 'negative'], ['58', 'female', 'typ_angina', '150', '283', 't', 'left_vent_hyper', '162', 'no', '1', 'up', '0', 'normal', 'negative'], ['50', 'female', 'non_anginal', '120', '219', 'f', 'normal', '158', 'no', '1.6', 'flat', '0', 'normal', 'negative'], ['58', 'female', 'non_anginal', '120', '340', 'f', 'normal', '172', 'no', '0', 'up', '0', 'normal', 'negative'], ['66', 'female', 'typ_angina', '150', '226', 'f', 'normal', '114', 'no', '2.6', 'down', '0', 'normal', 'negative'], ['43', 'male', 'asympt', '150', '247', 'f', 'normal', '171', 'no', '1.5', 'up', '0', 'normal', 'negative'], ['69', 'female', 'typ_angina', '140', '239', 'f', 'normal', '151', 'no', '1.8', 'up', '2', 'normal', 'negative'], ['59', 'male', 'asympt', '135', '234', 'f', 'normal', '161', 'no', '0.5', 'flat', '0', 'reversable_defect', 'negative'], ['44', 'male', 'non_anginal', '130', '233', 'f', 'normal', '179', 'yes', '0.4', 'up', '0', 'normal', 'negative'], ['42', 'male', 'asympt', '140', '226', 'f', 'normal', '178', 'no', '0', 'up', '0', 'normal', 'negative'], ['61', 'male', 'non_anginal', '150', '243', 't', 'normal', '137', 'yes', '1', 'flat', '0', 'normal', 'negative'], ['40', 'male', 'typ_angina', '140', '199', 'f', 'normal', '178', 'yes', '1.4', 'up', '0', 'reversable_defect', 'negative'], ['71', 'female', 'atyp_angina', '160', '302', 'f', 'normal', '162', 'no', '0.4', 'up', '2', 'normal', 'negative'], ['59', 'male', 'non_anginal', '150', '212', 't', 'normal', '157', 'no', '1.6', 'up', '0', 'normal', 'negative'], ['51', 'male', 'non_anginal', '110', '175', 'f', 'normal', '123', 'no', '0.6', 'up', '0', 'normal', 'negative'], ['65', 'female', 'non_anginal', '140', '417', 't', 'left_vent_hyper', '157', 'no', '0.8', 'up', '1', 'normal', 'negative'], ['53', 'male', 'non_anginal', '130', '197', 't', 'left_vent_hyper', '152', 'no', '1.2', 'down', '0', 'normal', 'negative']]

Do you know how can I get output like the following created by arff library( from Weka)?

Answer 1

You say "no external libraries", but can you at least cut-and-paste into your own code? You may find the source code to the arff module useful (200 lines, about 5.6 KB).

Edit:

You may find this format reference useful: http://weka.wikispaces.com/ARFF+%28stable+version%29

Edit2:

Just for fun, I wrote my own .arrf parser; it's almost as long as the WEKA code, but should be more readable - just six functions, a dispatch table, and a very modular class. You can iterate over a class instance to get each data row as a namedtuple.

See what you think:

from collections import namedtuple
from keyword import iskeyword
import re

def NotDone(msg):
    raise NotImplemented(msg)

def nominal(spec):
    """
    Create an ARFF nominal (enumerated) data type
    """
    spec = spec.lstrip("{ \t").rstrip("} \t")
    good_values = set(val.strip() for val in spec.split(","))

    def fn(s):
        s = s.strip()
        if s in good_values:
            return s
        else:
            raise ValueError("'{}' is not a recognized value".format(s))

    # patch docstring
    fn.__name__ = "nominal"
    fn.__doc__ = """
    ARFF nominal (enumerated) data type

    Legal values are {}
    """.format(sorted(good_values))
    return fn

def numeric(s):
    """
    Convert string to int or float
    """
    try:
        return int(s)
    except ValueError:
        return float(s)

field_maker = {
    "date":       (lambda spec: NotDone("date data type not implemented")),
    "integer":    (lambda spec: int),
    "nominal":    (lambda spec: nominal(spec)),
    "numeric":    (lambda spec: numeric),
    "string":     (lambda spec: str),
    "real":       (lambda spec: float),
    "relational": (lambda spec: NotDone("relational data type not implemented")),
}

def file_lines(fname):
    # lazy file reader; ensures file is closed when done,
    # returns lines without trailing spaces or newline
    with open(fname) as inf:
        for line in inf:
            yield line.rstrip()

def no_data_yet(*items):
    raise ValueError("AarfRow not fully defined (haven't seen a @data directive yet)")

def make_field_name(s):
    """
    Mangle string to make it a valid Python identifier
    """
    s = s.lower()                               # force to lowercase
    s = "_".join(re.findall("[a-z0-9]+", s))    # strip all invalid chars; join what's left with "_"
    if iskeyword(s) or re.match("[0-9]", s):    # if the result is a keyword or starts with a digit
        s = "f_"+s                              #   make it a safe field name
    return s  

class ArffReader:
    line_types = ["blank", "comment", "relation", "attribute", "data"]

    def __init__(self, fname):
        # get input file
        self.fname = fname
        self.lines = file_lines(fname)

        # prepare to read file header
        self.relation = '(not specified)'
        self.data_names = []
        self.data_types = []
        self.dtype = no_data_yet

        # read file header
        line_tests = [
            (getattr(self, "line_is_{}".format(item)), getattr(self, "line_do_{}".format(item)))
            for item in self.__class__.line_types
        ]
        for line in self.lines:
            for is_, do in line_tests:
                if is_(line):
                    done = do(line)
                    break
            if done:
                break

        # use header fields to build data type (and make it print as requested)
        class ArffRow(namedtuple('ArffRow', self.data_names)):
            __slots__ = ()
            def __str__(self):
                items = (getattr(self, field) for field in self._fields)
                return "({})".format(", ".join(repr(it) for it in items))
        self.dtype = ArffRow

    #
    # figure out input-line type
    #

    def line_is_blank(self, line):
        return not line

    def line_is_comment(self, line):
        return line.lower().startswith('%')

    def line_is_relation(self, line):
        return line.lower().startswith('@relation')

    def line_is_attribute(self, line):
        return line.lower().startswith('@attribute')

    def line_is_data(self, line):
        return line.lower().startswith('@data')

    #    
    # handle input-line type
    #    

    def line_do_blank(self, line):
        pass

    def line_do_comment(self, line):
        pass

    def line_do_relation(self, line):
        self.relation = line[10:].strip()

    def line_do_attribute(self, line):
        m = re.match(
            "^@attribute"           #   line starts with '@attribute'
            "\s+"                   #
            "("                     # name is one of:
                "(?:'[^']+')"       #   ' string in single-quotes '
                "|(?:\"[^\"]+\")"   #   " string in double-quotes "
                "|(?:[^ \t'\"]+)"   #   single_word_string (no spaces)
            ")"                     #
            "\s+"                   #
            "("                     # type is one of:
                "(?:{[^}]+})"       #   { set, of, nominal, values }
                "|(?:\w+)"          #   datatype
            ")"                     #
            "\s*"                   #
            "("                     # spec string
                ".*"                #   anything to end of line
            ")$",                   #
            line, flags=re.I)       #   case-insensitive
        if m:
            name, type_, spec = m.groups()
            self.data_names.append(make_field_name(name))
            if type_[0] == '{':
                type_, spec = 'nominal', type_
            self.data_types.append(field_maker[type_](spec))
        else:
            raise ValueError("failed parsing attribute line '{}'".format(line))

    def line_do_data(self, line):
        return True  # flag end of header

    #
    # make the class iterable
    #

    def __iter__(self):
        return self

    def next(self):
        """
        Return one data row at a time
        """
        data = next(self.lines).split(',')
        return self.dtype(*(fn(dat) for fn,dat in zip(self.data_types, data)))

and it can be used as

for row in ArffReader('mydata.arff'):
    print(row)

which results in

(63.0, 'male', 'typ_angina', 145.0, 233.0, 't', 'left_vent_hyper', 150.0, 'no', 2.3, 'down', 0.0, 'fixed_defect', 'negative')
(37.0, 'male', 'non_anginal', 130.0, 250.0, 'f', 'normal', 187.0, 'no', 3.5, 'down', 0.0, 'normal', 'negative')
(41.0, 'female', 'atyp_angina', 130.0, 204.0, 'f', 'left_vent_hyper', 172.0, 'no', 1.4, 'up', 0.0, 'normal', 'negative')
(56.0, 'male', 'atyp_angina', 120.0, 236.0, 'f', 'normal', 178.0, 'no', 0.8, 'up', 0.0, 'normal', 'negative')
(57.0, 'female', 'asympt', 120.0, 354.0, 'f', 'normal', 163.0, 'yes', 0.6, 'up', 0.0, 'normal', 'negative')
(57.0, 'male', 'asympt', 140.0, 192.0, 'f', 'normal', 148.0, 'no', 0.4, 'flat', 0.0, 'fixed_defect', 'negative')
(56.0, 'female', 'atyp_angina', 140.0, 294.0, 'f', 'left_vent_hyper', 153.0, 'no', 1.3, 'flat', 0.0, 'normal', 'negative')
(44.0, 'male', 'atyp_angina', 120.0, 263.0, 'f', 'normal', 173.0, 'no', 0.0, 'up', 0.0, 'reversable_defect', 'negative')
(52.0, 'male', 'non_anginal', 172.0, 199.0, 't', 'normal', 162.0, 'no', 0.5, 'up', 0.0, 'reversable_defect', 'negative')

The fields are also addressable by name, ie

for patient in ArffReader('mydata.arff'):
    print("{} year old {}".format(patient.age, patient.sex))

which gives

63.0 year old male
37.0 year old male
41.0 year old female
56.0 year old male
57.0 year old female
57.0 year old male
56.0 year old female
44.0 year old male
52.0 year old male

and you can see the file names by

>>> print(repr(patient))
ArffRow(age=63.0, sex='male', cp='typ_angina', trestbps=145.0, chol=233.0, fbs='t', restecg='left_vent_hyper', thalach=150.0, exang='no', oldpeak=2.3, slope='down', ca=0.0, thal='fixed_defect', f_class='negative')

Field names are as per the ARFF header, forced lowercase (and in the case of 'class' prepended with 'f_' because class is a Python keyword hence not usable as a field name).