teagrjohnson/few_shot_emotions_5epochs-20iter-300trainsize

SetFit with sentence-transformers/paraphrase-mpnet-base-v2

This is a SetFit model trained on the jakeazcona/short-text-multi-labeled-emotion-classification dataset that can be used for Text Classification. This SetFit model uses sentence-transformers/paraphrase-mpnet-base-v2 as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification.

The model has been trained using an efficient few-shot learning technique that involves:

1. Fine-tuning a Sentence Transformer with contrastive learning.
2. Training a classification head with features from the fine-tuned Sentence Transformer.

Model Details

Model Description

• Model Type: SetFit
• Sentence Transformer body: sentence-transformers/paraphrase-mpnet-base-v2
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 512 tokens
• Number of Classes: 6 classes
• Training Dataset: jakeazcona/short-text-multi-labeled-emotion-classification

Model Sources

• Repository: SetFit on GitHub
• Paper: Efficient Few-Shot Learning Without Prompts
• Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts

Model Labels

Label	Examples
2	'i could do was feel i felt thankful that her battle was over thankful that she was now in a place of serenity' 'i feel that i m indulging him at times nor does it help that when we started talking his approach was more friend zone friendly than an i want to date you approach' 'i was i might be buying stuff from there but i feel the clothes are too casual'
5	'Me too, except there’s no hookup. (I’m so lonely) ' 'Due to work, gotta stay legal. Edibles aren’t legal yet sadly. Fortunately not taking it for cancer though, just pain.' 'I don’t understand the point you’re making sorry'
3	'i feel most passionate about that arouse my emotions seem to be the things i need to learn something about my emotion tells me there is a need to grow in some direction' 'One of my favorite shows of all time.' 'i really feel like he will never love me he will never be affectionate because he doesnt love me'
4	'This is a restaurant in a hotel so it had a strange vibe.' 'i feel very shocked by how many people i talk to who havent seen this movie' "I thought you only get to be a wizard after you're 30yo and still virgin"
1	'i can t help but feeling weird when opening every closet in an apartment that somebody s still living in so i didn t' 'i want to stop taking it one day but also feel terrified that lots of feelings of anxiety panic will come flooding back' 'i have crossed over and i am on safe footing yet still feel this way fearful for the unknown shaky uncertain'
0	"??? Where the hell is this article getting it's data from..." 'i feel like i meet the most subtly obnoxious annoying people in the universe' 'This is so cruel. I literally feel physically sick.'

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("teagrjohnson/few_shot_emotions_5epochs-20iter-300trainsize")
# Run inference
preds = model("i feel stupid using this name")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	3	16.54	53

Label	Training Sample Count
0	47
1	28
2	86
3	29
4	23
5	87

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (5, 5)
• max_steps: -1
• sampling_strategy: oversampling
• num_iterations: 20
• body_learning_rate: (2e-05, 2e-05)
• head_learning_rate: 2e-05
• loss: CosineSimilarityLoss
• distance_metric: cosine_distance
• margin: 0.25
• end_to_end: False
• use_amp: False
• warmup_proportion: 0.1
• seed: 42
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0013	1	0.2647	-
0.0667	50	0.1911	-
0.1333	100	0.2146	-
0.2	150	0.1469	-
0.2667	200	0.1151	-
0.3333	250	0.1102	-
0.4	300	0.0334	-
0.4667	350	0.0291	-
0.5333	400	0.0057	-
0.6	450	0.0017	-
0.6667	500	0.0009	-
0.7333	550	0.0013	-
0.8	600	0.0006	-
0.8667	650	0.0005	-
0.9333	700	0.0009	-
1.0	750	0.0005	-
1.0667	800	0.0003	-
1.1333	850	0.0003	-
1.2	900	0.0003	-
1.2667	950	0.0002	-
1.3333	1000	0.0004	-
1.4	1050	0.0003	-
1.4667	1100	0.001	-
1.5333	1150	0.0002	-
1.6	1200	0.0001	-
1.6667	1250	0.0003	-
1.7333	1300	0.0003	-
1.8	1350	0.0002	-
1.8667	1400	0.0003	-
1.9333	1450	0.0002	-
2.0	1500	0.0002	-
2.0667	1550	0.0001	-
2.1333	1600	0.0002	-
2.2	1650	0.0001	-
2.2667	1700	0.0002	-
2.3333	1750	0.0001	-
2.4	1800	0.0001	-
2.4667	1850	0.0003	-
2.5333	1900	0.0001	-
2.6	1950	0.0001	-
2.6667	2000	0.0008	-
2.7333	2050	0.0001	-
2.8	2100	0.0001	-
2.8667	2150	0.0001	-
2.9333	2200	0.0001	-
3.0	2250	0.0002	-
3.0667	2300	0.0002	-
3.1333	2350	0.0001	-
3.2	2400	0.0001	-
3.2667	2450	0.0001	-
3.3333	2500	0.0001	-
3.4	2550	0.0001	-
3.4667	2600	0.0001	-
3.5333	2650	0.0001	-
3.6	2700	0.0001	-
3.6667	2750	0.0001	-
3.7333	2800	0.0001	-
3.8	2850	0.0001	-
3.8667	2900	0.0001	-
3.9333	2950	0.0001	-
4.0	3000	0.0004	-
4.0667	3050	0.0002	-
4.1333	3100	0.0001	-
4.2	3150	0.0001	-
4.2667	3200	0.0001	-
4.3333	3250	0.0001	-
4.4	3300	0.0001	-
4.4667	3350	0.0001	-
4.5333	3400	0.0001	-
4.6	3450	0.0001	-
4.6667	3500	0.0001	-
4.7333	3550	0.0001	-
4.8	3600	0.0001	-
4.8667	3650	0.0001	-
4.9333	3700	0.0001	-
5.0	3750	0.0001	-

Framework Versions

• Python: 3.10.12
• SetFit: 1.0.3
• Sentence Transformers: 2.2.2
• Transformers: 4.36.1
• PyTorch: 2.0.1+cu118
• Datasets: 2.15.0
• Tokenizers: 0.15.0

Citation

BibTeX

@article{https://doi.org/10.48550/arxiv.2209.11055,
    doi = {10.48550/ARXIV.2209.11055},
    url = {https://arxiv.org/abs/2209.11055},
    author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
    keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {Efficient Few-Shot Learning Without Prompts},
    publisher = {arXiv},
    year = {2022},
    copyright = {Creative Commons Attribution 4.0 International}
}