custom brown clusters

[cahuja1992]

Brown clusters are shipped as a part of resources. How can we use custom cluster

[adrienball]

Hi @cahuja1992, Using custom resources is possible but not well documented yet, it will be improved and documented in the coming months.

In the meantime, here is the process that you can follow to use your own resources in english:

• download the latest english resources (.tar.gz archive) on this page (https://github.com/snipsco/snips-nlu-language-resources/releases)
• untar the resource archive
• rename the resource name (snips_nlu_en) to your own custom resource name (for instance snips_nlu_custom_en) in the following places:
  • folder names (leave the version number)
  • "name" attribute in the metadata.json file
• add the name of your word clusters in the "word_clusters" list attribute of the metadata file. This should be the filename without the extension.
• add the word clusters file (.txt) in the word_clusters directory
• run pip install <path/to/resources_dir>, pointing to the resources directory that you customized
• run snips-nlu link <custom_resource_name> en --force. This will make the default english resources point to your customized version

At this point, your custom english resources are setup and you can reference your new word clusters in the nlu engine configuration (cf https://snips-nlu.readthedocs.io/en/latest/tutorial.html#the-snips-nlu-engine)

I hope this helps, Cheers

[adrienball]

Another way to use custom resources consists in manually loading resources, updating them, and then pass the customized resources to SnipsNLUEngine():

from snips_nlu import SnipsNLUEngine
from snips_nlu.resources import load_resources
from snips_nlu.default_configs.config_en import CONFIG

resources = load_resources("en")
custom_clusters_name = "my_custom_clusters"
resources["metadata"]["word_clusters"].append(custom_clusters_name)
resources["word_clusters"][custom_clusters_name] = {
    "hello": "word_cluster1",
    "world": "word_cluster2"
}

config = CONFIG
# ... Update the config to leverage the new word clusters
engine = SnipsNLUEngine(config=config, resources=resources)

[cahuja1992]

Thanks, it works.