mtyrrell/CPU_Conditional_Classifier

IKT_classifier_conditional_best

This model is a fine-tuned version of sentence-transformers/all-mpnet-base-v2 on the GIZ/policy_qa_v0_1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5371
• Precision Macro: 0.8714
• Precision Weighted: 0.8713
• Recall Macro: 0.8711
• Recall Weighted: 0.8712
• F1-score: 0.8712
• Accuracy: 0.8712

Model description

The model is a binary text classifier based on sentence-transformers/all-mpnet-base-v2 and fine-tuned on text sourced from national climate policy documents.

Intended uses & limitations

The classifier assigns a class of 'Unconditional' or 'Conditional' to denote the strength of commitments as portrayed in extracted passages from the documents. The intended use is for climate policy researchers and analysts seeking to automate the process of reviewing lengthy, non-standardized PDF documents to produce summaries and reports.

Due to inconsistencies in the training data, the classifier performance leaves room for improvement. The classifier exhibits reasonably good training metrics (F1 ~ 0.85), balanced between precise identification of true positive classifications (precision ~ 0.85) and a wide net to capture as many true positives as possible (recall ~ 0.85). When tested on real world unseen test data, the performance was subptimal for a binary classifier (F1 ~ 0.5). However, testing was based on a small out-of-sample dataset containing it's own inconsistencies. Therefore classification may prove more robust in practice.

Training and evaluation data

The training dataset is comprised of labelled passages from 2 sources:

• ClimateWatch NDC Sector data.
• IKI TraCS Climate Strategies for Transport Tracker implemented by GIZ and funded by theInternational Climate Initiative (IKI) of the German Federal Ministry for Economic Affairs and Climate Action (BMWK). Here we utilized the QA dataset (CW_NDC_data_Sector).

The combined datasetGIZ/policy_qa_v0_1 contains ~85k rows. Each row is duplicated twice, to provide varying sequence lengths (denoted by the values 'small', 'medium', and 'large', which correspond to sequence lengths of 60, 85, and 150 respectively - indicated in the 'strategy' column). This effectively means the dataset is reduced by 1/3 in useful size, and the 'strategy' value should be selected based on the use case. For this training, we utilized the 'medium' samples Furthermore, for each row, the 'context' column contains 3 samples of varying quality. The approach used to assess quality and select samples is described below.

The pre-processing operations used to produce the final training dataset were as follows:

1. Dataset is filtered based on 'medium' value in 'strategy' column (sequence length = 85).
2. For IKITracs, labels are assigned based on the presence of certain substrings ('_unc' or '_c') based on 'parameter' values which correspond to assessments of 'unconditional' or 'conditional' commitments by human annotaters.
3. For ClimateWatch, the 'QuestionText' field is searched for the terms 'unconditional' or 'conditional', and labels assigned accordingly.
4. If 'context_translated' is available and the 'language' is not English, 'context' is replaced with 'context_translated'. This results in the model being trained on English translations of original text samples.
5. The dataset is "exploded" - i.e., the text samples in the 'context' column, which are lists, are converted into separate rows - and labels are merged to align with the associated samples.
6. The 'match_onanswer' and 'answerWordcount' are used conditionally to select high quality samples (prefers high % of word matches in 'match_onanswer', but will take lower if there is a high 'answerWordcount')
7. Data is then augmented using sentence shuffle from the albumentations library (NLP methods insertion and substitution were also tried, but lowered the performance of the model and were therefore not included in the final training data). This is done to increase the number of training samples available for the Unconditional class from 774 to 1163. The end result is an equal sample per class breakdown of:

   • UNCONDITIONAL: 1163
   • CONDITIONAL: 1163

Training procedure

The model hyperparameters were tuned using optuna over 10 trials on a truncated training and validation dataset. The model was then trained over 5 epochs using the best hyperparameters identified.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 4.112924307850544e-05
• train_batch_size: 3
• eval_batch_size: 3
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 400.0
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Precision Macro	Precision Weighted	Recall Macro	Recall Weighted	F1-score	Accuracy
0.6658	1.0	698	0.7196	0.7391	0.7381	0.7102	0.7124	0.7028	0.7124
0.6301	2.0	1396	0.4965	0.8073	0.8075	0.8071	0.8069	0.8069	0.8069
0.5252	3.0	2094	0.5307	0.8300	0.8297	0.8279	0.8283	0.8279	0.8283
0.3513	4.0	2792	0.5261	0.8626	0.8627	0.8626	0.8627	0.8626	0.8627
0.2979	5.0	3490	0.5371	0.8714	0.8713	0.8711	0.8712	0.8712	0.8712

Framework versions

• Transformers 4.31.0
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3