camembert-base-finetuned-ICDCode_5

This model is a fine-tuned version of camembert-base on the None dataset. It has been trained on a corpus of death certificate. One ICDCode is given for a given cause of death or commorbidities. As it is an important task to be able to predict these ICDCode, I shave trained this model for 8 epochs on 400 000 death causes. Pre-processing of noisy data points was mandatory before tokenization. It allows us to get this accuracy. It achieves the following results on the evaluation set:

• Loss: 0.6574
• Accuracy: 0.8964
• F1: 0.8750
• Recall: 0.8964

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 50
• eval_batch_size: 50
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 8

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Recall
3.7466	1.0	4411	1.9448	0.7201	0.6541	0.7201
1.5264	2.0	8822	1.2045	0.8134	0.7691	0.8134
1.0481	3.0	13233	0.9473	0.8513	0.8149	0.8513
0.8304	4.0	17644	0.8098	0.8718	0.8427	0.8718
0.7067	5.0	22055	0.7352	0.8834	0.8574	0.8834
0.6285	6.0	26466	0.6911	0.8898	0.8659	0.8898
0.5779	7.0	30877	0.6641	0.8958	0.8741	0.8958
0.549	8.0	35288	0.6574	0.8964	0.8750	0.8964

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1