eval.py

#!/usr/bin/python
#coding: utf-8

'''
This script will download the dataset from the MLP project resources
then randomly shuffle it and split it in two to create a training and test set.

The student will have to train his model using the generated training set. He
will then evaluate the model performances at prediction on the test set.

The performance should be evaluated on binary cross-entropy loss function.
42ai - wwatkins@student.42.fr - 2017 '''

import io
import os
import sys
import random
import csv

def writeToCsv(path, content, labels):
    ''' write a given content in a csv file.
        arguments :
            `path` : (string)
                the path where we will save the csv file (relative or absolute path).
            `content` : (ndarray/list)
                the content to save in the csv file.
            `labels` : (ndarray/list)
                the labels of the csv columns.
    '''
    with io.open(path, 'w') as fd:
        writer = csv.writer(fd)
        if labels is not None:
            content.insert(0, labels)
        writer.writerows(content)

def splitList(content, cut):
    ''' splits an array in two.
        arguments :
            `content` : (ndarray/list)
                the array to split in two.
            `cut` : (float) in [0:1]
                the ratio by which we will split the array.
    '''
    c = int(len(content) * cut)
    return (content[:c], content[c:])

def splitDataset(path, cut=0.2, label=False, shuffle=False):
    ''' loads the dataset and splits it in two sets.
        arguments :
            `path` : (string)
                the path to the dataset to load (relative or absolute path).
            `cut` : (float) in [0:1]
                the ratio for the reparition of the two sets (0.2 means 20 percents
                for the test set and 80 percents for the training set).
            `label` : (boolean)
                wether the first line of the csv contains labels or not
            `shuffle` : (boolean)
                wether we shuffle the dataset before the split operation or not
    '''
    labels = None
    content = []
    csvfile = open(path, 'rb') if sys.version_info[0] < 3 else open(path, 'rt', encoding='utf-8')
    reader = csv.reader(csvfile, delimiter=',')
    for i, row in enumerate(reader):
        if i == 0 and label is True:
            labels = row
            continue
        content.append(row)
    csvfile.close()
    if shuffle is True:
        random.shuffle(content)
    filename = path[:path.rfind('.')]
    testset, trainingset = splitList(content, float(cut))
    writeToCsv(filename + '_test.csv', testset, labels)
    writeToCsv(filename + '_training.csv', trainingset, labels)


dataset_url = 'https://projects.intra.42.fr/uploads/document/document/464/data.csv'
filepath = './data.csv'
os.system('curl -s -o {1} {0}'.format(dataset_url, filepath))
splitDataset(filepath, cut=0.25, label=False, shuffle=True)