Dockerfile and example for Facebook Research's fastText.
If you'd like to use a cluster manager to deploy fastText-docker, we have configurations for:
See more in deployments
The quickest way to see the fastText classification tutorial with fastText-docker is:
docker pull xebxeb/fasttext-docker
mkdir -p /tmp/data && mkdir -p /tmp/result
docker run --rm -v /tmp/data:/data -v /tmp/result:/result -it xebxeb/fasttext-docker ./classification-example.sh
NOTE: if you ran the above on macOS, the data & results are going to be on your Docker Machine VM. Use a path in /Users/${USER}/ if you want to map to your local system. Or do something like: docker-machine ssh `docker-machine active` ls /tmp/data to see the files in your VM.
There are two Dockerfiles, including:
- Dockerfile all-in-one, used for development purposes. Includes the fastText binary, the entire source repository and Python dependencies.
- Dockerfile.binary just for executing the fastText binary. The fasttext binary is the container's entrypoint.
If you'd like to use the published prebuilt images, you can pull them from DockerHub. NOTE: the latest will always be the devel tag.
docker pull xebxeb/fasttext-docker
docker pull xebxeb/fasttext-docker:devel
docker pull xebxeb/fasttext-docker:binary
To build the devel Docker image, simply do a make after cloning or:
docker build -t fasttext .
Building the docker image will create the fasttext binary as well as clone the repository -- all in the root of the container.
There is no entrypoint for the devel container and any of the examples in the fastText repository will work.
The development container is meant to be interactive, so the best way to use it is probably with a mounted volume and bash.
$ mkdir -p /tmp/data && mkdir -p /tmp/result
$ docker run --rm -it -v /tmp/data:/data -v /tmp/result:/result fasttext /bin/bash
# ./fasttext
usage: fasttext <command> <args>
The commands supported by fasttext are:
supervised train a supervised classifier
test evaluate a supervised classifier
predict predict most likely label
skipgram train a skipgram model
cbow train a cbow model
print-vectors print vectors given a trained model
# ./classification-example.sh
Resolving googledrive.com (googledrive.com)... 216.58.194.33, 2607:f8b0:4000:802::2001
Connecting to googledrive.com (googledrive.com)|216.58.194.33|:443... connected.
HTTP request sent, awaiting response... 302 Moved Temporarily
....
dbpedia_csv/
dbpedia_csv/classes.txt
dbpedia_csv/test.csv
dbpedia_csv/train.csv
dbpedia_csv/readme.txt
make: Nothing to be done for `opt'.
Read 32M words
Progress: 50.2% words/sec/thread: 1833592 lr: 0.049821 loss: 0.141374 eta: 0h0m
You get the idea... it's a full interactive shell with a mounted volume.
NOTE be sure to use absolute paths in your local mount arguments! And if you are on macOS, be sure that your path is within /Users/ -- otherwise you will map to a path on your Docker Machine VM. See classification-example.sh for an example.
If you'd just like a pre-built binary of fastText, you can build the binary simply by doing:
make binary
(and modifying the Makefile to your image name if you'd like)
You will likely want to mount a volume with Docker in order to use the binary container because it has an entrypoint of the fasttext binary. For example:
docker run --rm -v /var/path/to/data:/data -v /var/path/to/results:/results test "/result/dbpedia.bin" "/data/dbpedia.test"
docker run --rm -v /var/path/to/data:/data -v /var/path/to/results:/results predict "/result/dbpedia.bin" "/data/dbpedia.test" > "data/dbpedia.test.predict"
See classification-example.sh for an example using the devel tag. Simply replace that with binary and remove the ./fastText argument to achieve the same result.