INSTRUCTIONS

instructions to run in Mac OS, Linux or WSL inside Windows

Install conda

link below for 64-bit linux:

wget https://repo.anaconda.com/archive/Anaconda3-2020.02-Linux-x86_64.sh

Run after download to install:

bash Anaconda3-2020.02-Linux-x86_64.sh

Import conda environments

First env is for deepCT and my custom scripts

conda-env create -n saved-env-deepCT -f=conda_envs/saved-env-deepCT.yml

Second env is for pyserini

conda-env create -n saved-env-pyserini -f=conda_envs/saved-env-pyserini.yml

Create DB from patent files

Activate deepCT conda env

conda activate saved-env-deepCT

Get conda env's python path to be sure we are using the right python executable

echo $CONDA_PREFIX

will be something like /Users/giorgossideris/opt/anaconda3/envs/saved-env-deepCT

Create db

IMPORTANT (before runnning create-db.py, edit path of patents inside file to correspond to the correct location in your filesystem) sample files are included in the repo, if you want to use those, change path in file to sample_patents

<PYTHON_PATH_FROM_PREVIOUS_STEP>/bin/python ./create-db.py

keep the limit and fields you used in mind to use later for parameters eg. if you used limit 0 and fields claims,abstract, these 2 values will be used below as EXPERIMENT_NAME, in this case 0-claims,abstract

Change to pyserini env and calculate tf-idf scores

Change to pyserini env

conda activate saved-env-pyserini

Get conda env's python path to be sure we are using the right python executable

echo $CONDA_PREFIX

Run pyserini inverted index creation

<PYTHON_PATH_FROM_PREVIOUS_STEP>/bin/python -m pyserini.index --input jsonl/<EXPERIMENT_NAME> --collection JsonCollection --generator DefaultLuceneDocumentGenerator --index indexes/<EXPERIMENT_NAME> --stemmer=none --threads 1 --storePositions --storeDocvectors --storeRaw

Calculate tf idf scores with pyserini

<PYTHON_PATH_FROM_PREVIOUS_STEP>/bin/python pyserini_tfIdf.py

Run deepCT training and prediction

Change to deepCT env

conda activate saved-env-deepCT

Get conda env's python path to be sure we are using the right python executable

echo $CONDA_PREFIX

Run deepCT training

before running deepCT training download https://storage.googleapis.com/bert_models/2020_02_20/uncased_L-12_H-768_A-12.zip, unzip and copy bert_model.ckpt.data-00000-of-00001 to folder bert-base-uncased

<PYTHON_PATH_FROM_PREVIOUS_STEP>/bin/python DeepCT-master/run_deepct.py --data_dir=output/<EXPERIMENT_NAME>/train.docterm_recall --vocab_file=bert-base-uncased/vocab.txt --bert_config_file=bert-base-uncased/bert_config.json --init_checkpoint=bert-base-uncased/bert_model.ckpt --output_dir=output/<EXPERIMENT_NAME>/train --do_train=true --task_name=marcodoc --num_train_epochs=3.0 --train_batch_size=16

Create file needed for deepCT prediction

<PYTHON_PATH_FROM_PREVIOUS_STEP>/bin/python ./create-file-for-edit.py

ΙMPORTANT: remove all lines with empty content before moving on to the next step

Run deepCT prediction

<PYTHON_PATH_FROM_PREVIOUS_STEP>/bin/python DeepCT-master/run_deepct.py --task_name=marcotsvdoc --do_train=false --do_eval=false --do_predict=true --data_dir=output/<EXPERIMENT_NAME>/edit.tsv --vocab_file=bert-base-uncased/vocab.txt --bert_config_file=bert-base-uncased/bert_config.json --init_checkpoint=output/<EXPERIMENT_NAME>/train/model.ckpt-0 --max_seq_length=128 --train_batch_size=16 --learning_rate=2e-5 --num_train_epochs=3.0 --output_dir=output/<EXPERIMENT_NAME>/predict

Turn float-point term wegihts into TF-like index weights

python DeepCT-master/scripts/bert_term_sample_to_json.py --output_format=json output/inf-description/edit.tsv  output/inf-description/predict/test_results.tsv output/inf-description/jsonl/predict/predict.json 100

example result

{
  "id": "EP-1089487",
  "contents": "an an an an an an an an an an an an an an an an an an an an an an an an an an an an an an an an an and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and and method method method method method method method method method method method method method method for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for for ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering ciphering traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic traffic exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged exchanged in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in in both both both both both both both both both directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions directions between between between between between between between between between between between between a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite satellite / / / / / / / / / / / / / telephone telephone telephone telephone telephone telephone telephone ground station station station station station station station station station station station station station station station station station station station station station station station station station station station station station network network network network network network network network network network network network network network network network network network network network network network network network network network network network network network network network network network network using using using using using using using using using using using orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial orbitial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . buffer buffer buffer buffer buffer buffer buffer buffer buffer buffer buffer buffer buffer buffer is is is is is is is is is is is is is is is at at at at at at at at at at at at at at at at at at either either either either either either either either either either either either the the the the the the the the the the the the the the the the the the the the the the the the the the the the the or or or or or or or or or or or or or or or or or or or or or or or or or or or to to to to to to to to to to to to to to to to to to to to to to to to to to to to to to deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering deciphering bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits bits output output output output output output output output from duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex duplex algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm algorithm generated generated generated generated generated generated generated generated generated are are are decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher decipher later later later later later later later later later later later later later later - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - be be be be be be be be be be be be be be be be received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received received information information information information information information information information information information information information information information information information block block block block block block block block block block block block block block block block block block block block block block block block block block block block block block block each each each each each each each each each each each each each each each each each each each each each each each call call call call call call call call call call call call call call call call call call up up up up nearest nearest nearest nearest nearest nearest nearest nearest nearest integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer integer number number number number number number number number number number number number number number number number of of of of of of of of of of of of of of of of of of of of"
}

(same as jsonl created with pyseriniCreateInvertedIndexHelpFile)

Create index with pyserini

same as before

Run queries to index

python run_queries_to_anserini_index.py (parameters, filepaths defined in the file)

Evaluation of test runs with a qrels file

in cygwin: ./trec_eval-master/trec_eval.exe -m map 3973_qrels.txt query_results/300_topics_set3_EN_description.res