SetFit

This is a SetFit model that can be used for Text Classification. A SVC instance is used for classification.

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

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

Model Details

Model Description

Model Type: SetFit
Classification head: a SVC instance
Maximum Sequence Length: 256 tokens
Number of Classes: 2 classes

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
SUBJ	'Gone are the days when they led the world in recession-busting' 'Who so mean that he will not himself be taxed, who so mindful of wealth that he will not favor increasing the popular taxes, in aid of these defective children?' 'That state has sixty-two counties and sixty cities … In addition there are 932 towns, 507 villages, and, at the last count, 9,600 school districts … Just try to render efficient service … amid the diffused identities and inevitable jealousies of, roughly, 11,000 independent administrative officers or boards!'
OBJ	'Is this a warning of what’s to come?' 'This unique set of circumstances has brought PCL back into focus as the safe haven of choice for global players seeking somewhere to stash their cash.' 'Socialists believe that, if everyone cannot have something, no one shall.'

Label

Examples

SUBJ

'Gone are the days when they led the world in recession-busting'
'Who so mean that he will not himself be taxed, who so mindful of wealth that he will not favor increasing the popular taxes, in aid of these defective children?'
'That state has sixty-two counties and sixty cities … In addition there are 932 towns, 507 villages, and, at the last count, 9,600 school districts … Just try to render efficient service … amid the diffused identities and inevitable jealousies of, roughly, 11,000 independent administrative officers or boards!'

OBJ

'Is this a warning of what’s to come?'
'This unique set of circumstances has brought PCL back into focus as the safe haven of choice for global players seeking somewhere to stash their cash.'
'Socialists believe that, if everyone cannot have something, no one shall.'

Evaluation

Metrics

Label	F1
all	0.7526

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("SOUMYADEEPSAR/Setfit_subj_SVC")
# Run inference
preds = model("That can happen again.")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	3	35.9834	97

Label	Training Sample Count
OBJ	117
SUBJ	124

Training Hyperparameters

batch_size: (8, 8)
num_epochs: (5, 5)
max_steps: -1
sampling_strategy: oversampling
num_iterations: 20
body_learning_rate: (1e-05, 1e-05)
head_learning_rate: 1e-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
0.0008	1	0.3862	-
0.0415	50	0.4092	-
0.0830	100	0.3596	-
0.1245	150	0.2618	-
0.1660	200	0.2447	-
0.2075	250	0.263	-
0.2490	300	0.2583	-
0.2905	350	0.3336	-
0.3320	400	0.2381	-
0.3734	450	0.2454	-
0.4149	500	0.259	-
0.4564	550	0.2083	-
0.4979	600	0.2437	-
0.5394	650	0.2231	-
0.5809	700	0.0891	-
0.6224	750	0.1164	-
0.6639	800	0.0156	-
0.7054	850	0.0394	-
0.7469	900	0.0065	-
0.7884	950	0.0024	-
0.8299	1000	0.0012	-
0.8714	1050	0.0014	-
0.9129	1100	0.0039	-
0.9544	1150	0.0039	-
0.9959	1200	0.001	-
1.0373	1250	0.0007	-
1.0788	1300	0.0003	-
1.1203	1350	0.001	-
1.1618	1400	0.0003	-
1.2033	1450	0.0003	-
1.2448	1500	0.0014	-
1.2863	1550	0.0003	-
1.3278	1600	0.0003	-
1.3693	1650	0.0001	-
1.4108	1700	0.0004	-
1.4523	1750	0.0003	-
1.4938	1800	0.0008	-
1.5353	1850	0.0002	-
1.5768	1900	0.0005	-
1.6183	1950	0.0002	-
1.6598	2000	0.0004	-
1.7012	2050	0.0001	-
1.7427	2100	0.0002	-
1.7842	2150	0.0002	-
1.8257	2200	0.0002	-
1.8672	2250	0.0003	-
1.9087	2300	0.0001	-
1.9502	2350	0.0002	-
1.9917	2400	0.0001	-
2.0332	2450	0.0003	-
2.0747	2500	0.0002	-
2.1162	2550	0.0001	-
2.1577	2600	0.0001	-
2.1992	2650	0.0004	-
2.2407	2700	0.0002	-
2.2822	2750	0.0001	-
2.3237	2800	0.0005	-
2.3651	2850	0.0002	-
2.4066	2900	0.0003	-
2.4481	2950	0.0001	-
2.4896	3000	0.0001	-
2.5311	3050	0.0001	-
2.5726	3100	0.0001	-
2.6141	3150	0.0002	-
2.6556	3200	0.0001	-
2.6971	3250	0.0002	-
2.7386	3300	0.0002	-
2.7801	3350	0.0001	-
2.8216	3400	0.0001	-
2.8631	3450	0.0001	-
2.9046	3500	0.0001	-
2.9461	3550	0.0	-
2.9876	3600	0.0002	-
3.0290	3650	0.0001	-
3.0705	3700	0.0	-
3.1120	3750	0.0001	-
3.1535	3800	0.0001	-
3.1950	3850	0.0001	-
3.2365	3900	0.0001	-
3.2780	3950	0.0001	-
3.3195	4000	0.0001	-
3.3610	4050	0.0001	-
3.4025	4100	0.0	-
3.4440	4150	0.0001	-
3.4855	4200	0.0001	-
3.5270	4250	0.0001	-
3.5685	4300	0.0001	-
3.6100	4350	0.0002	-
3.6515	4400	0.0001	-
3.6929	4450	0.0001	-
3.7344	4500	0.0	-
3.7759	4550	0.0	-
3.8174	4600	0.0001	-
3.8589	4650	0.0001	-
3.9004	4700	0.0001	-
3.9419	4750	0.0	-
3.9834	4800	0.0001	-
4.0249	4850	0.0001	-
4.0664	4900	0.0001	-
4.1079	4950	0.0001	-
4.1494	5000	0.0	-
4.1909	5050	0.0	-
4.2324	5100	0.0	-
4.2739	5150	0.0	-
4.3154	5200	0.0001	-
4.3568	5250	0.0001	-
4.3983	5300	0.0001	-
4.4398	5350	0.0	-
4.4813	5400	0.0001	-
4.5228	5450	0.0	-
4.5643	5500	0.0001	-
4.6058	5550	0.0001	-
4.6473	5600	0.0001	-
4.6888	5650	0.0	-
4.7303	5700	0.0001	-
4.7718	5750	0.0001	-
4.8133	5800	0.0001	-
4.8548	5850	0.0	-
4.8963	5900	0.0	-
4.9378	5950	0.0	-
4.9793	6000	0.0001	-

Framework Versions

Python: 3.10.12
SetFit: 1.0.3
Sentence Transformers: 2.7.0
Transformers: 4.40.1
PyTorch: 2.2.1+cu121
Datasets: 2.19.0
Tokenizers: 0.19.1

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}
}

SOUMYADEEPSAR
/

Setfit_subj_SVC