iati-climate-classifier

This model is a fine-tuned version of alex-miller/ODABert on a subset of the alex-miller/iati-policy-markers dataset.

It achieves the following results on the evaluation set:

• Loss: 0.2377
• Accuracy: 0.9138
• F1: 0.9165
• Precision: 0.8889
• Recall: 0.9458

Model description

This model has been trained to identify climate mitigation and climate adaptation project titles and/or descriptions. It returns "0" for projects with no climate component, and "1" for projects with adaptation or mitigation as principal objectives.

Training procedure

Code to subset the dataset and train the model is available here.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-06
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 7

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss	Precision	Recall
0.4992	1.0	876	0.8921	0.8978	0.2831	0.8530	0.9475
0.2706	2.0	1752	0.9038	0.9057	0.2446	0.8881	0.9241
0.2494	3.0	2628	0.9095	0.9114	0.2370	0.8927	0.9309
0.2393	4.0	3504	0.9112	0.9140	0.2385	0.8863	0.9435
0.2306	5.0	4380	0.9124	0.9152	0.2380	0.8870	0.9452
0.229	6.0	5256	0.2405	0.9121	0.9152	0.8836	0.9492
0.2255	7.0	6132	0.2377	0.9138	0.9165	0.8889	0.9458

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2