SamLowe/all-MiniLM-L6-v2-go_emotions-classifier-onnx

Overview

This is a multi-label, multi-class linear classifer for emotions that works with sentence-transformers/all-MiniLM-L6-v2, having been trained on the go_emotions dataset.

Labels

The 28 labels from the go_emotions dataset are:

['admiration', 'amusement', 'anger', 'annoyance', 'approval', 'caring', 'confusion', 'curiosity', 'desire', 'disappointment', 'disapproval', 'disgust', 'embarrassment', 'excitement', 'fear', 'gratitude', 'grief', 'joy', 'love', 'nervousness', 'optimism', 'pride', 'realization', 'relief', 'remorse', 'sadness', 'surprise', 'neutral']

Metrics (exact match of labels per item)

This is a multi-label, multi-class dataset, so each label is effectively a separate binary classification. Evaluating across all labels per item in the go_emotions test split the metrics are shown below.

Optimising the threshold per label to optimise the F1 metric, the metrics (evaluated on the go_emotions test split) are:

• Precision: 0.384
• Recall: 0.438
• F1: 0.397

Weighted by the relative support of each label in the dataset, this is:

• Precision: 0.443
• Recall: 0.552
• F1: 0.484

Using a fixed threshold of 0.5 to convert the scores to binary predictions for each label, the metrics (evaluated on the go_emotions test split, and unweighted by support) are:

• Precision: 0.551
• Recall: 0.211
• F1: 0.261

Metrics (per-label)

This is a multi-label, multi-class dataset, so each label is effectively a separate binary classification and metrics are better measured per label.

Optimising the threshold per label to optimise the F1 metric, the metrics (evaluated on the go_emotions test split) are:

	f1	precision	recall	support	threshold
admiration	0.529	0.499	0.563	504	0.25
amusement	0.733	0.672	0.807	264	0.20
anger	0.394	0.363	0.429	198	0.15
annoyance	0.293	0.252	0.350	320	0.15
approval	0.292	0.345	0.254	351	0.20
caring	0.320	0.270	0.393	135	0.15
confusion	0.291	0.276	0.307	153	0.15
curiosity	0.366	0.307	0.454	284	0.15
desire	0.317	0.269	0.386	83	0.15
disappointment	0.159	0.127	0.212	151	0.10
disapproval	0.306	0.341	0.277	267	0.20
disgust	0.405	0.412	0.398	123	0.20
embarrassment	0.364	0.414	0.324	37	0.35
excitement	0.296	0.232	0.408	103	0.15
fear	0.496	0.576	0.436	78	0.40
gratitude	0.793	0.787	0.798	352	0.30
grief	0.323	0.200	0.833	6	0.45
joy	0.402	0.341	0.491	161	0.15
love	0.640	0.679	0.605	238	0.30
nervousness	0.263	0.333	0.217	23	0.70
optimism	0.433	0.453	0.414	186	0.20
pride	0.429	0.500	0.375	16	0.50
realization	0.177	0.159	0.200	145	0.10
relief	0.182	0.182	0.182	11	0.40
remorse	0.541	0.500	0.589	56	0.30
sadness	0.461	0.467	0.455	156	0.20
surprise	0.302	0.299	0.305	141	0.15
neutral	0.620	0.505	0.803	1787	0.30

The thesholds are stored in thresholds.json.

Use with ONNXRuntime

The input to the model is called logits, and there is one output per label. Each output produces a 2d array, with 1 row per input row, and each row having 2 columns - the first being a proba output for the negative case, and the second being a proba output for the positive case.

# Assuming you have embeddings from all-MiniLM-L6-v2 for the input sentences
# E.g. produced from sentence-transformers such as:
#      huggingface.co/sentence-transformers/all-MiniLM-L6-v2
#      or from an ONNX version E.g. huggingface.co/Xenova/all-MiniLM-L6-v2

print(embeddings.shape)  # E.g. a batch of 1 sentence
> (1, 384)

import onnxruntime as ort

sess = ort.InferenceSession("path_to_model_dot_onnx", providers=['CPUExecutionProvider'])

outputs = [o.name for o in sess.get_outputs()]  # list of labels, in the order of the outputs
preds_onnx = sess.run(_outputs, {'logits': embeddings})
# preds_onnx is a list with 28 entries, one per label,
# each with a numpy array of shape (1, 2) given the input was a batch of 1

print(outputs[0])
> surprise
print(preds_onnx[0])
> array([[0.97136074, 0.02863926]], dtype=float32)

# load thresholds.json and use that (per label) to convert the positive case score to a binary prediction 

Commentary on the dataset

Some labels (E.g. gratitude) when considered independently perform very strongly, whilst others (E.g. relief) perform very poorly.

This is a challenging dataset. Labels such as relief do have much fewer examples in the training data (less than 100 out of the 40k+, and only 11 in the test split).

But there is also some ambiguity and/or labelling errors visible in the training data of go_emotions that is suspected to constrain the performance. Data cleaning on the dataset to reduce some of the mistakes, ambiguity, conflicts and duplication in the labelling would produce a higher performing model.