Comands to run to get experiments working:

• python3 -m venv venv
• source venv/bin/activate
• git clone https://github.com/UFAL-DSG/tgen.git
• git clone https://github.com/tuetschek/e2e-metrics.git
• git clone https://github.com/thushv89/attention_keras.git
• mv e2e-metrics e2e_metrics
• pip3 install -r requirements.txt
• pip3 install -r e2e_metrics/requirements.txt
• pip3 install git+https://github.com/ufal/pytreex These next commands are needed to use the e2e_metrics without change working directory to the e2e_metrics directory
• sed -i '' -e 's/from pycocotools/from e2e_metrics.pycocotools/g' e2e_metrics/measure_scores.py
• sed -i '' -e 's/from pycocoevalcap.eval/from e2e_metrics.pycocoevalcap.eval/g' e2e_metrics/measure_scores.py
• sed -i '' -e 's/from metrics.pymteval/from e2e_metrics.metrics.pymteval/g' e2e_metrics/measure_scores.py

The comands setup the data from the e2e-dataset for use in this thesis

• cd tgen/e2e-challenge/input
• ./convert.py -a name,near -n ../../../e2e-dataset/trainset.csv train
• ./convert.py -a name,near -n -m ../../../e2e-dataset/devset.csv devel
• cd ../../..

To train the Seq2seq model:

• python3 train_seq2seq.py -c new_configs/model_configs/seq2seq_model.yaml

To test the vanilla beam search:

• python3 get_results.py -c new_configs/vanilla_beam_search.yaml

To train beam manipulator model:

• python3 get_results.py -c new_configs/setup_bm_traindata.yaml
• python3 train_beam_manipulator.py -c new_configs/model_configs/bm_model.yaml

To test beam manipulator:

• python3 get_results.py -c new_configs/bm_test.yaml

To view the bleu scores of the experiments:

• python3 get_results_bleu_scores.py For information on running further experiments contact jameshargreaves12@gmail.com