Feature/term table (#3)

* makes more sense to preload BERTje before copying app.py

* updated aethel version

* return serialized proof and lexical phrases in response

* parse errors should be (imo) 500

Dockerfile

@@ -10,7 +10,7 @@ RUN git clone https://github.com/konstantinosKokos/spindle.git
 WORKDIR /spindle
 
 # Install aethel
-RUN pip install git+https://github.com/konstantinosKokos/aethel@795f34046b7970a28e0e2491ba23dea5e716f1d2
+RUN pip install git+https://github.com/konstantinosKokos/aethel@41eab8fb178a197cdf8de738b68e386f07e6e4f5
 
 # Install PyTorch and its dependencies
 RUN pip3 install torch==1.12.0 opt_einsum --extra-index-url https://download.pytorch.org/whl/cpu
@@ -23,6 +23,9 @@ RUN pip3 install --no-index \
 
 RUN pip3 install transformers==4.20.1 six Flask
 
+# Download BERTje model ahead of time
+RUN python -c 'from transformers import pipeline; pipeline("fill-mask", model="GroNLP/bert-base-dutch-cased")'
+
 # Copy data files
 COPY atom_map.tsv data/atom_map.tsv
 COPY bert_config.json data/bert_config.json
@@ -39,8 +42,5 @@ ENV FLASK_APP=app.py
 # Shows print logs from our server in the container logs.
 ENV PYTHONUNBUFFERED=1
 
-# Download BERTje model ahead of time
-RUN python -c 'from transformers import pipeline; pipeline("fill-mask", model="GroNLP/bert-base-dutch-cased")'
-
 # Run the Flask server
 CMD ["flask", "run", "--host=0.0.0.0"]

app.py

@@ -1,6 +1,7 @@
 from flask import Flask, abort, jsonify, request
 import json
 from inference import InferenceWrapper
+from aethel.mill.serialization import serial_proof_to_json, serialize_proof
 from aethel.utils.tex import sample_to_tex
 import logging
 
@@ -44,16 +45,34 @@ def handle_request():
     log.info("Analysis complete!")
     log.info("Results: %s", results)
 
+    if len(results) < 1:
+        log.error("Got no results")
+        abort(500)
+
+    analysis = results[0]
+    # spindle will store an exception value in the proof variable, at least in some failure modes
+    if isinstance(analysis.proof, Exception):
+        log.error("Error in analysis", exc_info=analysis.proof)
+        abort(500)
+
     try:
-        tex_from_sample = sample_to_tex(results[0])
+        tex_from_sample = sample_to_tex(analysis)
     except:
-        log.error("Failed to convert result to TeX.")
-        abort(400)
+        log.exception("Failed to convert result to TeX.")
+        abort(500)
+        return  # not necessary given abort, but helps type-checker understand that we leave the function here
 
     log.info("TeX conversion successful.")
     log.info("TeX: %s", tex_from_sample)
 
-    response = {"results": tex_from_sample}
+    # prepare json-ready version of proof and lexical phrases
+    proof = serial_proof_to_json(serialize_proof(analysis.proof))
+    lexical_phrases = [phrase.json() for phrase in analysis.lexical_phrases]
+
+    response = dict(
+        tex=tex_from_sample,
+        proof=proof,
+        lexical_phrases=lexical_phrases)
     return jsonify(response)