tomaarsen HF staff commited on
Commit
9ec985d
1 Parent(s): dca5006

Add training script etc.

Browse files
Files changed (1) hide show
  1. README.md +64 -0
README.md CHANGED
@@ -59,6 +59,70 @@ The data fields are the same among all splits.
59
  |---------|----:|---------:|---:|
60
  |tomaarsen/setfit-absa-semeval-laptops|2358|49|654|
61
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  ### Citation Information
63
 
64
  ```bibtex
 
59
  |---------|----:|---------:|---:|
60
  |tomaarsen/setfit-absa-semeval-laptops|2358|49|654|
61
 
62
+ ### Training ABSA models using SetFit ABSA
63
+
64
+ To train using this dataset, first install the SetFit library:
65
+
66
+ ```bash
67
+ pip install setfit
68
+ ```
69
+
70
+ And then you can use the following script as a guideline of how to train an ABSA model on this dataset:
71
+
72
+ ```python
73
+ from setfit import AbsaModel, AbsaTrainer, TrainingArguments
74
+ from datasets import load_dataset
75
+ from transformers import EarlyStoppingCallback
76
+
77
+ # You can initialize a AbsaModel using one or two SentenceTransformer models, or two ABSA models
78
+ model = AbsaModel.from_pretrained("sentence-transformers/all-MiniLM-L6-v2")
79
+
80
+ # The training/eval dataset must have `text`, `span`, `polarity`, and `ordinal` columns
81
+ dataset = load_dataset("tomaarsen/setfit-absa-semeval-laptops")
82
+ train_dataset = dataset["train"]
83
+ eval_dataset = dataset["validation"]
84
+
85
+ args = TrainingArguments(
86
+ output_dir="models",
87
+ use_amp=True,
88
+ batch_size=256,
89
+ eval_steps=50,
90
+ save_steps=50,
91
+ load_best_model_at_end=True,
92
+ )
93
+
94
+ trainer = AbsaTrainer(
95
+ model,
96
+ args=args,
97
+ train_dataset=train_dataset,
98
+ eval_dataset=eval_dataset,
99
+ callbacks=[EarlyStoppingCallback(early_stopping_patience=5)],
100
+ )
101
+ trainer.train()
102
+
103
+ metrics = trainer.evaluate(eval_dataset)
104
+ print(metrics)
105
+
106
+ trainer.push_to_hub("tomaarsen/setfit-absa-laptops")
107
+ ```
108
+
109
+ You can then run inference like so:
110
+ ```python
111
+ from setfit import AbsaModel
112
+
113
+ # Download from Hub and run inference
114
+ model = AbsaModel.from_pretrained(
115
+ "tomaarsen/setfit-absa-laptops-aspect",
116
+ "tomaarsen/setfit-absa-laptops-polarity",
117
+ )
118
+
119
+ # Run inference
120
+ preds = model([
121
+ "Boots up fast and runs great!",
122
+ "The screen shows great colors.",
123
+ ])
124
+ ```
125
+
126
  ### Citation Information
127
 
128
  ```bibtex