Deepcubes services wrapper

Flask wrappers for different models from deepcubes package

Services, scripts and experiment section

Just several examples of usage

Vera Live Dialog API

/train

POST query with config field as json string representaion:

{
	"lang": string,  # now just "rus" or "eng"
	"not_understand_label": string,  # any string that corresponds to `not understand` label
	"labels_settings": [
		{
			"label": string,  # unique label name
			"patterns": [string],  # list with regexps, example: ["нет", "нет.*"]
			"generics": [string],  # list with generics, possible:
                                               # ["yes", "no", "repeat", "no"questions"]
			"intent_phrases": [string],  # list with intent phrases for ML
		},
		...
	]
}

Returns json string with model_id (int).

/predict

POST query with model_id (int) field (returned by /train) and query (string) field as user input text. Additionall labels ([array representation]) field can be specified, in this case model returns probabilities only for specified labels.

Returns collection of labels sorted decreasingly according probabilities.

[
	{
		"label": string,
		"proba": float
	},
	...
]

Embedder service

/get_vectors

POST query with tokens ([[string, string, ...], [string, string, ...], ...]) fields.

Returns json string with vectors field, lists of floats with embedding vectors components.

Authors

• Dmitry Ischenko
• Svyatoslav Nevyantsev