Fixed ordering issue, updated Dockerfile

Dockerfile

@@ -1,6 +1,7 @@
 FROM python:3.6-slim-buster as final
 
 RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential\
     gcc \
     cmake \
     git \
@@ -17,9 +18,12 @@ COPY requirements.txt requirements.txt
 COPY alpaca_client/requirements.txt alpaca_client/requirements.txt
 COPY alpaca_server/requirements.txt alpaca_server/requirements.txt
 
+# increases pip timemout limit to avoid errors downloading pytorch
 # 1 fetch small torch - for GPU support use a runtime image from hub.docker.com/r/pytorch/pytorch
 # 2 install requirements for both django and model server
 # 3 download spacy and distilbert data
+ENV PIP_DEFAULT_TIMEOUT 100
+
 RUN pip install --no-cache-dir \
     torch==1.7.1+cpu \
     -f https://download.pytorch.org/whl/torch_stable.html && \
@@ -43,8 +47,6 @@ RUN npm install --no-cache && \
     npm run build
 
 
-
-
 FROM final
 
 WORKDIR /alpaca
@@ -67,6 +69,7 @@ ENV [email protected]
 ENV DATABASE_URL=sqlite:////db.sqlite3
 ENV ALLOW_SIGNUP=False
 ENV DEBUG=True
+ENV SECRET_KEY teste
 
 # create django admin/migrations, start alpaca-serving, start django
 COPY startup.sh startup.sh

alpaca_server/alpaca_model/kerasAPI/acquisition.py

@@ -35,7 +35,8 @@ def get_mnlp(self, x_train, model, preprocess, num_tokens, batch_size=50):
             scores = np.array(scores)
             norm_scores = scores/wordslen
             assert len(norm_scores) == len(batch_x)
-            probscores.extend(list(norm_scores[wordslen.argsort()[::-1]]))
+            reverse_idx = wordslen.argsort()[::-1].argsort()
+            probscores.extend(list(norm_scores[reverse_idx]))
 
         assert len(new_datapoints) == len(probscores)
         probs[new_datapoints] = np.array(probscores)

alpaca_server/alpaca_model/pytorchAPI/active_learning/acquisition.py

@@ -62,6 +62,7 @@ def get_mnlp(self, dataset, model, num_instances, batch_size = 50):
             wordslen = data['wordslen']
             charslen = data['charslen']
             sort_info = data['sort_info']
+            inverse_sort = np.asarray(sort_info).argsort()
 
             score = model.decode(words, chars, caps, wordslen, charslen, mask, usecuda = self.usecuda,
                                  score_only = True)