emotion_model_improved

This model is a fine-tuned version of xlm-roberta-large on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2881
• Macro F1: 0.5947
• Micro F1: 0.6896
• Accuracy: 0.8522
• F1 Anger: 0.8051
• Precision Anger: 0.7756
• Recall Anger: 0.8368
• F1 Anticipation: 0.3591
• Precision Anticipation: 0.3484
• Recall Anticipation: 0.3705
• F1 Disgust: 0.7122
• Precision Disgust: 0.6203
• Recall Disgust: 0.8360
• F1 Fear: 0.7222
• Precision Fear: 0.6506
• Recall Fear: 0.8115
• F1 Joy: 0.8601
• Precision Joy: 0.8641
• Recall Joy: 0.8561
• F1 Sadness: 0.7075
• Precision Sadness: 0.6030
• Recall Sadness: 0.8558
• F1 Surprise: 0.2393
• Precision Surprise: 0.3305
• Recall Surprise: 0.1875
• F1 Trust: 0.2643
• Precision Trust: 0.2242
• Recall Trust: 0.3217
• F1 Love: 0.6566
• Precision Love: 0.7855
• Recall Love: 0.5640
• F1 Optimism: 0.7413
• Precision Optimism: 0.7730
• Recall Optimism: 0.7122
• F1 Pessimism: 0.4745
• Precision Pessimism: 0.3367
• Recall Pessimism: 0.8032
• Positive Predictions Pct: 25.8683
• Positive Labels Pct: 21.7367

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: 8e-06
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Macro F1	Micro F1	Accuracy	F1 Anger	Precision Anger	Recall Anger	F1 Anticipation	Precision Anticipation	Recall Anticipation	F1 Disgust	Precision Disgust	Recall Disgust	F1 Fear	Precision Fear	Recall Fear	F1 Joy	Precision Joy	Recall Joy	F1 Sadness	Precision Sadness	Recall Sadness	F1 Surprise	Precision Surprise	Recall Surprise	F1 Trust	Precision Trust	Recall Trust	F1 Love	Precision Love	Recall Love	F1 Optimism	Precision Optimism	Recall Optimism	F1 Pessimism	Precision Pessimism	Recall Pessimism	Positive Predictions Pct	Positive Labels Pct
0.6834	1.0	72	0.4816	0.2295	0.4570	0.6345	0.5297	0.3603	1.0	0.0	0.0	0.0	0.4483	0.2889	1.0	0.0	0.0	0.0	0.5649	0.3936	1.0	0.4936	0.3277	1.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.4884	0.3238	0.9937	0.0	0.0	0.0	45.3927	21.9208
0.4738	2.0	144	0.3507	0.4607	0.6320	0.7951	0.7069	0.5593	0.9604	0.0	0.0	0.0	0.6359	0.4807	0.9388	0.4906	0.3694	0.7302	0.8185	0.7592	0.8877	0.6246	0.4692	0.9336	0.0	0.0	0.0	0.0	0.0	0.0	0.6323	0.5235	0.7980	0.7368	0.6486	0.8529	0.4223	0.2951	0.7422	33.7556	21.9208
0.3445	3.0	216	0.3131	0.5845	0.6929	0.8585	0.7807	0.8	0.7623	0.3463	0.4804	0.2707	0.7273	0.6953	0.7624	0.7003	0.7637	0.6465	0.8477	0.8140	0.8843	0.7385	0.6994	0.7822	0.1839	0.2051	0.1667	0.2020	0.1360	0.3924	0.6915	0.6374	0.7557	0.7637	0.7268	0.8046	0.4473	0.3785	0.5467	24.1394	21.9208
0.3076	4.0	288	0.3035	0.5792	0.6867	0.8561	0.7741	0.7691	0.7792	0.3486	0.3904	0.3149	0.7255	0.6826	0.7741	0.6789	0.7818	0.6	0.8348	0.7760	0.9033	0.7201	0.7044	0.7365	0.2316	0.2340	0.2292	0.1949	0.1364	0.3418	0.6912	0.6792	0.7036	0.7587	0.7018	0.8256	0.4125	0.3882	0.44	24.0158	21.9208
0.2836	5.0	360	0.2969	0.6002	0.7045	0.8648	0.7859	0.7927	0.7792	0.3462	0.4122	0.2983	0.7387	0.6950	0.7882	0.7364	0.6926	0.7860	0.8543	0.8160	0.8964	0.7339	0.7137	0.7552	0.2735	0.2319	0.3333	0.2190	0.1756	0.2911	0.6983	0.6779	0.7199	0.7653	0.7440	0.7878	0.4508	0.3927	0.5289	23.8366	21.9208
0.27	6.0	432	0.2930	0.6238	0.6993	0.8541	0.8007	0.7733	0.8302	0.3858	0.4167	0.3591	0.7349	0.6604	0.8282	0.7578	0.7316	0.7860	0.8506	0.8412	0.8601	0.7366	0.7045	0.7718	0.4051	0.5161	0.3333	0.2334	0.1477	0.5570	0.7170	0.6466	0.8046	0.7745	0.7596	0.7899	0.4650	0.3395	0.7378	26.5991	21.9208
0.2587	7.0	504	0.2888	0.6137	0.6969	0.8525	0.7948	0.7756	0.8151	0.3526	0.3697	0.3370	0.7387	0.6667	0.8282	0.7348	0.7704	0.7023	0.8528	0.8406	0.8653	0.7384	0.6776	0.8112	0.3505	0.3469	0.3542	0.2185	0.1403	0.4937	0.7166	0.6524	0.7948	0.7703	0.7370	0.8067	0.4831	0.3532	0.7644	26.7474	21.9208
0.248	8.0	576	0.2865	0.6177	0.6960	0.8520	0.7923	0.7691	0.8170	0.3802	0.3596	0.4033	0.7329	0.6712	0.8071	0.7379	0.7716	0.7070	0.8560	0.8219	0.8929	0.7317	0.7096	0.7552	0.3738	0.3390	0.4167	0.2259	0.1444	0.5190	0.7233	0.6486	0.8176	0.7671	0.7261	0.8130	0.4734	0.3548	0.7111	26.7598	21.9208
0.2404	9.0	648	0.2865	0.6219	0.7087	0.8617	0.7959	0.7900	0.8019	0.3913	0.3850	0.3978	0.7417	0.6811	0.8141	0.7489	0.6902	0.8186	0.8579	0.8361	0.8808	0.7390	0.6831	0.8050	0.3542	0.3542	0.3542	0.2368	0.1812	0.3418	0.7254	0.7721	0.6840	0.7747	0.7525	0.7983	0.475	0.3455	0.76	25.5547	21.9208

Framework versions

• Transformers 4.48.2
• Pytorch 2.3.1.post300
• Datasets 2.2.1
• Tokenizers 0.21.0