Programming language detector written in Python
A programming language detector written in Python using Scikit-Learn. Uses a Random Forest Classifier and it was trained to correctly identify a total of 50 programming languages, which are the following: Ada, AppleScript, AWK, BBC BASIC, C, C++, C#, Clojure, COBOL, Common Lisp, D, Elixir, Erlang, Forth, Fortran, Go, Groovy, Haskell, Icon, J, Java, JavaScript, Julia, Kotlin, LiveCode, Lua, Maple, MATLAB, Objective C, OCaml, Oz, Perl, PHP, PL-I, PowerShell, Prolog, Python, R, Racket, REXX, Ring, Ruby, Rust, Scala, Scheme, Swift, Tcl, UNIX Shell, VBScript and Visual-Basic .NET.
In order to have a large enough dataset for the above languages, the Roseta Code Dataset was used for training. Below are some metrics that were produced with 10-Fold Cross Validation in order to determine the performance of the trained classifier:
| Accuracy | Precision | Recall | F1 |
|---|---|---|---|
| 93.93% | 94.77% | 92.75% | 93.51% |
Note: in order to produce the above results, 80% of the dataset was used training and the other 20% for calculating the performance of the classifier.
To get the code up and running on your local machine, simply follow the following instructions.
First, you need to download scikit-learn (version 0.19 and newer) using the following command:
pip3 install -U scikit-learn
Note: make sure python3 (version 3.5 and newer) is installed.
To download the source code of this project use the following command:
git clone https://github.com/vsakkas/prog-lang-detector.git
And to enter the directory of the downloaded project, simply type:
cd prog-lang-detector
To train the classifier, simply run the following command:
python3 src/prog_lang_detector.py --train <dataset>
In the above command, <dataset> needs to be folder. The provided argument must end with a / and it must contain at least 50 directories, one for each of the languages to be used for training.
Note: running the above command will generate the following pickle files: dataset.pkl, tfidf.pkl, nmf.pkl, train.pkl, test.pkl, classifier.pkl. The last file contains the trained classifier. This file along with tfidf.pkl and nmf.pkl are required in order for the --predict command to work.
Finally, to use the trained classifier to predict what language a file us, use the following command:
python3 src/prog_lang_detector.py --predict <file>
The command above will simply print the predicted language for the given file.
Note: the extension of the provided file is not taken under consideration when trying to predict what programming language the code of the file is. Only the file’s content is used.
This project is licensed under the MIT License - see the LICENSE file for details