Command line tool to perform sentiment analysis written in C++ with Docker.
yasa can be built as a Docker image. The Dockerfile located in the project root directory takes care of all the steps to build the image. You only have to cd to the root directory and run:
$ docker build -t yasa .
This command will download all the needed dependencies, it will execute all the tests and finally Docker image called yasa will be built.
The Docker image can be directly executed with the Docker run command.
The following command is an example of usage:
$ docker run \
--mount type=bind,source=<path-to-yasa.db>,target=/yasa.db \
--mount type=bind,source=<path-to-pos-dir>,target=/<pos-dir>,readonly \
--mount type=bind,source=<path-to-neg-dir>,target=/<neg-dir>,readonly \
--rm --tty yasa train -p <pos-dir> -n <neg-dir>
After trained, the model is able to classify a text file using the following command:
$ docker run \
--mount type=bind,source=<path-to-yasa.db>,target=/yasa.db \
--mount type=bind,source=<path-to-file-to-classify>,target=/<file-to-classify>,readonly \
--rm --tty yasa classify -f <file-to-classify>
The yasa.db file must exists (you can create it with $ touch yasa.db), moreover files and directories used must be bound to the container, which can be achieved with Docker volumes or Docker bind mounts. The training phase will update this file yasa.db, which can be stored and used later for the classification phase or for another training.
For example in order to train the underneath model with all the text files in the directories example/pos and example/neg, you can use the following command (it can take about 5 minutes):
$ docker run \
--mount type=bind,source="$PWD"/yasa.db,target=/yasa.db \
--mount type=bind,source="$PWD"/example/pos,target=/pos,readonly \
--mount type=bind,source="$PWD"/example/neg,target=/neg,readonly \
--rm --tty yasa train -p pos -n neg
If you want to classify e.g. the example/neg.txt file with the model trained just before, you can use the following command:
$ docker run \
--mount type=bind,source="$PWD"/yasa.db,target=/yasa.db \
--mount type=bind,source="$PWD"/example/neg.txt,target=/neg.txt,readonly \
--rm --tty yasa classify -f neg.txt
These last three command can be executed simply by running the shell script run_example.sh.
yasa can also be built in the local system. We have only tested it in Ubuntu 18.04. You need the following dependencies (which are easily installable in Ubuntu with apt-get):
- CMake >= 3.10.2
- Google Test 1.8
- Google Mock 1.8
- SQLite 3.22 (libsqlite-dev in Ubuntu)
- clang-tidy >= 9
- gcov 7.4.0
- lcov 1.13
To build yasa and run all the tests with a coverage report:
$ mkdir build
$ cd build
$ cmake -j$(nproc) -DCMAKE_BUILD_TYPE=DEBUG ..
...
$ make -j$(nproc) RunAllTests
...
$ tests/RunAllTests
...
$ lcov --capture --directory . --output-file coverage.info
...
$ lcov --remove coverage.info \
'/usr/include/*' \
'/usr/local/include/*' \
$PWD'/tests/*' \
--quiet --output-file coverage.info
$ lcov --list coverage.info
...
To build yasa and install it in the system:
$ cmake -j$(nproc) -DCMAKE_BUILD_TYPE=Release ..
$ make -j$(nproc) yasa
$ cd src
$ sudo make install
To check if yasa is correctly installed run:
$ yasa --version
The main advantage with a local build is that you don't have to manage the data with Docker volumes or similar, you can directly run yasa on the local files and directories. For example:
$ yasa train -n example/neg -p example/pos
$ yasa classify -f example/pos.txt
...
$ yasa classify -f example/neg.txt