Modelpublic

Multilingual-Perspectivist-NLU/irony_nl_Netherlands

sourceHugging Facemitupdated 3y agoView on Hugging Face

0likes5downloads

Model Card

ironynlNetherlands

This model is a fine-tuned version of roberta-base on part of the MultiPICo dataset. It achieves the following results on the evaluation set:

—Loss: 0.0030
—Accuracy: 0.5738
—Precision: 0.2735
—Recall: 0.6275
—F1: 0.3810

Model description

The model is trained considering the annotation of annotators from the Netherlands only, on instances in Dutch. The Dutch annotations from these annotators are aggregated using majority voting and then used to train the model.

Training and evaluation data

The model has been trained on the annotation from annotators from the Netherlands from the MultiPICo dataset (instances in Dutch). The data has been randomly split into a train and a validation set.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 5e-06
—trainbatchsize: 16
—evalbatchsize: 64
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
0.0045	1.0	46	0.0043	0.7008	0.1562	0.0980	0.1205
0.0041	2.0	92	0.0042	0.7336	0.2083	0.0980	0.1333
0.004	3.0	138	0.0041	0.6393	0.2239	0.2941	0.2542
0.004	4.0	184	0.0039	0.4754	0.2484	0.7451	0.3725
0.0037	5.0	230	0.0037	0.5697	0.2589	0.5686	0.3558
0.0034	6.0	276	0.0036	0.4713	0.2532	0.7843	0.3828
0.0036	7.0	322	0.0036	0.5656	0.2477	0.5294	0.3375
0.0037	8.0	368	0.0036	0.5	0.2482	0.6863	0.3646
0.0031	9.0	414	0.0035	0.5287	0.2612	0.6863	0.3784
0.0034	10.0	460	0.0036	0.5369	0.25	0.6078	0.3543
0.0029	11.0	506	0.0035	0.5410	0.256	0.6275	0.3636
0.003	12.0	552	0.0032	0.5492	0.2677	0.6667	0.3820
0.0028	13.0	598	0.0030	0.5164	0.2752	0.8039	0.41
0.0026	14.0	644	0.0028	0.5574	0.2857	0.7451	0.4130
0.0024	15.0	690	0.0030	0.5820	0.2893	0.6863	0.4070
0.0023	16.0	736	0.0030	0.5738	0.2735	0.6275	0.3810

Framework versions

—Transformers 4.34.1
—Pytorch 2.0.1+cu117
—Datasets 2.14.5
—Tokenizers 0.14.1