This paper applies for new Python user but familiar with one or more other programming languages as well as having a solid experience on modularized software development.
The candidates are expected to demonstrate a fast learning skill to master python development environment setup in less than 6 hours working time and the ability to grasp correct utilization of nltk Python library.
- In telecommunication industry, automation is more and more important even vital to the success of fast response and iterative delivery. Furthermore, when a big number of software candidate issues are reported, the techniques of automatic sorting is still not well defined over this industry.
- The candidates are expected to design a program which utilizes appropriate Python libraries to train a classifier with given "training data" and generate results on "test data". After that the candidates can calculate the accuracy and prepare a description on how to improve it.
- During the training step, key
training-data[0]["issues"]['fields']['project']['key']is regarded as catalog title and the features (word list) should be extracted fromtraining-data[0]['fields']['description']]. In other words, the candidate should training classifier with given project names (as "TONTC", "PREM") and description section (a word list in JSON). - During the test action,
- Candidates can upload codes to their own repo on github, gitcafe, bitbuckets.
- Alternatively, if candidate is more familiar with IPython, it is appreciated if s/he would send back the IPython-NB format file in email or share the link.
- Training data and test data have the same format in this paper.
- Training data JSON format:
data[0]['total'] # the total number of issue records in this JSON object
data[0]["issues"] # the list of issues in this JSON object
data[0]["issues"][i]['fields']['project']['key'] # the project name for given issue record
data[0]["issues"][i]['fields']['summary'] # The given word list to feed to classifier
Below is the data structure. Here the author only list the data to process and ignore the other columns.
JSON
| total | issues |
|---|---|
| 220 | list |
issues
| fields |
|---|
| list |
fields
| description | project |
|---|---|
| The text list to process | list |
project: ['key'] is the project name as catalog title.
Project names in training-data are expected to be utilized to supervise the classifier but project names in test-data are only allowed to be used when checking results.
| key |
|---|
| TONTC |
- Candidates are expected to describe what happened to below operation on Linux or Cygwin(Linux environment simulator) to get a warm-up on this problem first.
[mawu.NANL-MAWU] ➤ curl -D- -b ../cookies/cache-2015-0822.maxwu -X POST -H "Content-Type: application/json" --data '{ "jql":"reporter in (zhangsan, lisi, wangwu, maliu) AND created >= 2015-08-15 AND created <2015-08-22 AND type = Bug","maxResults":10000,"fields":["id","key"]}' "http://jira.calix.local/rest/api/2/search" |grep '"total":[0-9]*' -o |tr ":" " " |awk '{print "Total"=$2,($2>=15)? "Yes": "No" }' % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 6997 0 6787 100 210 5414 167 0:00:01 0:00:01 --:--:-- 5455 42 Yes
[2015-08-23 12:44.25] ~/devel/GoJira/4interview [mawu.NANL-MAWU] ➤
- Given a small training set, the accurancy is expected to be higher than 60%.
- Candidates can directly pick up default algorithm in nltk or any other official Python package on PyPi service.
- Candidates shall prepare to present the logics and improvement points.
- Candidates shall consider if this technique can be applied to test case selection while planning test project w/in shared source code branches to different HW module products.
- It would be an advantage if candidate can compare multiple algorithm.