Dnnc config arg

annotators/dnnc_intents/Dockerfile

@@ -0,0 +1,25 @@
+FROM deeppavlov/base-gpu:0.17.5
+
+RUN pip install git+https://github.com/deeppavlov/[email protected]
+
+WORKDIR /base/DeepPavlov
+
+WORKDIR /src
+RUN mkdir common
+
+COPY annotators/dnnc_intents/requirements.txt ./requirements.txt
+RUN pip install -r requirements.txt
+
+RUN python -m spacy download en_core_web_sm
+
+ARG SERVICE_PORT
+ENV SERVICE_PORT=$SERVICE_PORT
+ARG CONFIG
+ENV CONFIG=$CONFIG
+ARG CLASSES
+ENV CLASSES=$CLASSES
+
+COPY annotators/dnnc_intents/ ./
+COPY common/ common/
+
+CMD gunicorn --workers=1 server:app -b 0.0.0.0:${SERVICE_PORT} --timeout=1200 --preload

annotators/dnnc_intents/Readme.md

@@ -0,0 +1,10 @@
+# DNNC intent classifier
+## Description
+Classifier that classifies DNNC intent in the few-shot mode. Default version is the logistic regression. Howeve, with config classifier_roberta.json  entailment-trained roberta is also supported
+## Input/Output
+**Input**
+Batch of last utterances
+**Output**
+For each utterance - dictionary {probable class: probability}
+## Dependencies
+As stated in requirements.txt. And if you use entailment-trained roberta, file utils_roberta.py also is useful

annotators/dnnc_intents/classifier.json

@@ -0,0 +1,75 @@
+{
+  "chainer": {
+    "in": ["x"],
+    "pipe": [
+       {
+        "class_name": "stream_spacy_tokenizer",
+                "in": [
+                    "x"
+                ],
+                "id": "my_tokenizer",
+                "lemmas": false,
+                "out": "token_lemmas",
+                "spacy_model": "en_core_web_sm",
+                "support_load_path": "{DOWNLOADS_PATH}/logreg_files/support_dataset.json"
+            },
+            {
+                "ref": "my_tokenizer",
+                "in": [
+                    "token_lemmas"
+                ],
+                "out": [
+                    "text_lem"
+                ]
+            },
+            {
+                "class_name": "fasttext",
+                "in": [
+                    "token_lemmas"
+                ],
+                "load_path": "{DOWNLOADS_PATH}/embeddings/fasttext/{LANGUAGE}.bin",
+                "mean": true,
+                "out": [
+                    "text_vector"
+                ]
+            },
+            {
+                "in": [
+                    "text_vector"
+                ],
+                "out": [
+                    "y_pred_probas"
+                ],
+                "class_name": "sklearn_component",
+                "main": true,
+                "save_path": "{DOWNLOADS_PATH}/logreg_files/model.pkl",
+                "load_path": "{DOWNLOADS_PATH}/logreg_files/model.pkl",
+                "model_class": "sklearn.linear_model:LogisticRegression",
+                "infer_method": "predict_proba",
+                "C": 10,
+                "penalty": "l2"
+            }
+    ],
+    "out": ["y_pred_probas"]
+  },
+  "metadata": {
+    "variables": {
+      "ROOT_PATH": "~/.deeppavlov",
+      "DOWNLOADS_PATH":"{ROOT_PATH}/downloads",
+      "LOGREG_PATH": "{DOWNLOADS_PATH}/logreg_files",
+      "LANGUAGE":"en"
+    },
+    "imports":["utils"],
+    "download": [
+            {
+                "url": "http://files.deeppavlov.ai/embeddings/fasttext/{LANGUAGE}.bin",
+                "subdir": "{DOWNLOADS_PATH}/embeddings/fasttext"
+            },
+            {
+                "url": "https://files.deeppavlov.ai/dream_data/dnnc_intents/logreg_files.tar.gz",
+                "subdir": "{LOGREG_PATH}"
+            }
+
+    ]
+  }
+}

annotators/dnnc_intents/classifier_roberta.json

@@ -0,0 +1,61 @@
+{
+  "chainer": {
+    "in": ["text"],
+    "pipe": [
+      {
+        "class_name": "dnnc_preparer",
+        "in": ["text"],
+        "out": ["texts","dataset"]
+      },
+      {
+        "class_name": "dnnc_pairgenerator",
+        "in": ["texts", "dataset"],
+        "out": ["x", "x_support", "x_populated", "y_support"],
+        "bidirectional": false
+      },
+      {
+        "class_name": "torch_transformers_preprocessor",
+        "in": ["x_populated", "x_support"],
+        "out": ["bert_features"],
+        "vocab_file": "{BASE_MODEL}",
+        "do_lower_case": true,
+        "max_seq_length": 128
+      },
+      {
+        "class_name": "torch_transformers_classifier_batch1",
+        "main": true,
+        "in": ["bert_features"],
+        "out": ["simmilarity_scores"],
+        "n_classes": 2,
+        "return_probas": true,
+        "pretrained_bert": "{BASE_MODEL}",
+        "save_path": "{MODEL_PATH}/model",
+        "load_path": "{MODEL_PATH}/model",
+        "is_binary": "{BINARY_CLASSIFICATION}"
+      },
+      {
+        "class_name": "dnnc_proba2labels",
+        "is_binary": "{BINARY_CLASSIFICATION}",
+        "in": ["simmilarity_scores", "x", "x_populated", "x_support", "y_support"],
+        "out": ["y_pred"],
+        "confidence_threshold": 0.0
+      }
+    ],
+    "out": ["y_pred"]
+  },
+  "metadata": {
+    "variables": {
+      "ROOT_PATH": "~/.deeppavlov",
+      "MODEL_PATH": "{ROOT_PATH}/models/fewshot/roberta_nli_mrpc_1_10",
+      "BINARY_CLASSIFICATION": true,
+      "BASE_MODEL": "roberta-base"
+    },
+    "imports":["utils_roberta"],
+    "download": [
+      {
+        "url": "http://files.deeppavlov.ai/v1/classifiers/fewshot/roberta_nli_mrpc_1_10.tar.gz",
+        "subdir": "{MODEL_PATH}"
+      }
+    ]
+  }
+}