Text Classification GoEmotions

This model is a fine-tuned version of roberta-large on the go_emotions dataset.

Model description

At first, 4 epochs of training with a learning rate of 5e-5 was performed on the roberta-large model. After that, the weights were loaded in a new environment and another epoch of training was done (this time with a learning rate of 2e-5). As the performance decreased in the fifth epoch, further training was discontinued.

After the 4th epoch, the model achieved a macro-F1 score of 53% on the test set, but the fifth epoch reduced the performance. The model on commit "5b532728cef22ca9e9bacc8ff9f5687654d36bf3" attains the following scores on the test set:

Accuracy: 0.4271236410539893
Precision: 0.5101494353184485
Recall: 0.5763722014150806
macro-F1: 0.5297380709491947

Load this specific version of the model using the syntax below:

import os
from transformers import AutoTokenizer, AutoModelForSequenceClassification

os.environ["TOKENIZERS_PARALLELISM"] = "FALSE"

model_name = "tasinhoque/text-classification-goemotions"
commit = "5b532728cef22ca9e9bacc8ff9f5687654d36bf3"
tokenizer = AutoTokenizer.from_pretrained(model_name, revision=commit)

model = AutoModelForSequenceClassification.from_pretrained(
    model_name, 
    num_labels=n_emotion, 
    problem_type="multi_label_classification", 
    revision=commit
)

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: 5e-05 (2e-5 in the 5th epoch)
train_batch_size: 128
eval_batch_size: 128
seed: 42 (only in the 5th epoch)
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
No log	1.0	340	0.0884	0.3782	0.4798	0.4643	0.4499
0.1042	2.0	680	0.0829	0.4093	0.4766	0.5272	0.4879
0.1042	3.0	1020	0.0821	0.4202	0.5103	0.5531	0.5092
0.0686	4.0	1360	0.0830	0.4327	0.5160	0.5556	0.5226
No log	5.0	1700	0.0961	0.4521	0.5190	0.5359	0.5218

Framework versions

Transformers 4.20.1
Pytorch 1.12.0
Datasets 2.1.0
Tokenizers 0.12.1

tasinhoque
/

text-classification-goemotions