few-shot-priming 0.3.937

Analyzing priming effects in a few shot setting environment

Exploring LLM Priming Strategies for Few-Shot Stance Classification

How does priming affect prompt-based learing? This project aims at analyzing this effect in stance classification. We train a stance classifier on the ibm stance classification dataset by fine-tuning an Alpaca model with a prompt and analyzing how does the selection of the few shots used in the prompt affect the performance of the model.

1. To evaluate the prompt-fine-tuning, run the following command

• Hyperparamter optimization

python scripts/run_prompt_fine_tuning.py --validate --optimize 

• T evaluate the prompt fine-utning approach using best hyperparameters in the config file.

python scripts/run_prompt_fine_tuning.py --validate  

2. To evaluate the prompting approaches (prompt) setup run

python scripts/run_prompting.py --validate --optimize 

3. To optmize DeBERTa using the hyperparameters in the config file, run the following

python scripts/optimize_baseline.py 

4. run the majority baseline

 python baseline.py --vast --majority --offline

5. Run prompting or prompt-fine tuning approaches with anaylze k as folows

python scripts/run_prompt_fine_tuning.py --analyze-k  

The ks or topic-counts are stored in conf.yaml and the analysis will be run for multiple seeds and the performance will be averaged and saved in the corresponding entry in /bigwork/nhwpajjy. To perform an analysis for one k instnance you can give it as a parameter using --k to produce the results for one seed you can specify the seed using --seed

2. Store the path of the produced csv file in conf.yaml under analyze-k

The results of the experiments will be logged to your home directory. The parameters can be saved in config.yaml

Priming Sampling strategies

To run an experiment with a priming sampling strategy use the parameter --similar-examples. This will retrieve priming examples tailored toward the test instance. The priming examples are selected using Sentence-Transformers --sentence-transformer. Example,

python scripts/run_prompting.py --validate --topic-similar --sentence-transformer

To run an experiment with a diversification sampling strategy use the parameter --diverse-examples. This will takes precomputed cluster centers of the training set as few shots. The number of clusters is then the few shot count provided in the configuration.

Topic Priming

Topic-priming examples are sampled using a topic similarity, which relies on sentence transformer. the similarities can be used to apply the right sampling strategy.

1. To compute the similarities between all the validation and training arguments run the following

python scripts/run_develop_similarity_measure.py --compute-similarity --ibmsc

You have to specify the model used to compute the similarity to be --sentence-transformer

To load similarities from the code you can use the

similarities = load_similarities("ibmsc", "validation", similarity)

which returns a dictionary of dictionary where the indices of the first dictionary are the test indices and the indeices of the nested dictionary for each test index are the indices of the training set with the values being the similarity scores.

To find similar or diverse arguments for an argument in the validation or test set, you can use

examples = sample_diverse_examples(experiment, experiment_type, few_shot_size)

similar can be used for sample_similar

examples = sample_similar_examples(test_istance_index, similarities, df_training, few_shot_size)

Similar examples for alll datasets are cached for efficiency and can be created by running the following commands

python argument_sampling/sampling_strategies.py" --ibmsc

to generate similar examples while using percentiles similarity thresholds you can use

python argument_sampling/sampling_strategies.py" "--ibmsc" --percentiles 10

Stance Priming

Stance-priming approaches use stance similarity to retrieve examples with similar stance to an input instance. To train the contrastive similarity measure, run

python scripts/traing_contrastive_learning.py --vast --lr --epochs --batch-size --output-path-model  

The best parameters are also saved in the config file, which will be used as default for each dataset.

To optimize the contrastive similarity measure on the validation set using the hyperparamters in the script, run.

python scripts/traing_contrastive_learning.py --vast --optimize  --results-path "$DATA_PATH/contrastive_learning/models/optimization-contrastive-learning-ibmsc.tsv"

To generate the similarity matrix using the contrastive similarity measure use

python scripts/generate_stance_priming_samples.py --vast --similarity-matrix 

To generate the stance priming samples using most similar , run the following

python scripts/generate_stance_priming_samples.py --vast --different-topics 

to generate similar examples while using percentiles similarity thresholds you can use

python scripts/generate_stance_priming_samples.py" "--ibmsc" --percentiles 10 --different-topics

Diversification strategy

To sample diverse examples, we use ward hierarchical clustering algorithm to cluster the trianing examples into $k$ cluster. The center of each cluster is then taken as an example. To find the diverse examples, we use the following command

python scripts/run_sample_diverse_examples --vast --validate --ibmsc 

This will precompute the cluster centers for k in [2, 4, 8, 16, 32, 64] To sample diverse examples, you can use the following command

examples = sample_diverse_examples(experiment, experiment_type, few_shot_size) where experiment is ibmsc or vast and experiment_type is validation or test.

Analysis

Mainly there are three types of analyses implemented on the prompting and instruction fine-tuning approaches. For this the results for all ks and the specific model and counts should be in the config file

Few shot size or training topic count effect on priming approaches

python script/run_visualize_over_k_performance.py --k --prompting --prompt-fine-tuning

2. Run the signifance tests. For this you need to store the predictions of the models using --path-predictions for all seeds

python notebooks/signifiance_test.py

3. Run the prime anaylis for stance priming you can run

 python "$CODE_PATH/scripts/run_prompting.py"  "--alpaca-7b" "--ibmsc" --offline --vllm --analyze-prime-similarity  \
 --similar-examples --path-similar-examples "$DATA_PATH/sampling-strategies/ibmsc-similar-stance-examples.tsv"

3. Run the prime anaylis for topic priming you can run

 python "$CODE_PATH/scripts/run_prompting.py"  "--alpaca-7b" "--ibmsc" --offline --vllm --analyze-topic-similarity  \
 --similar-examples --path-similar-examples "$DATA_PATH/sampling-strategies/ibmsc-similar-stance-examples.tsv"