Emotionallity Detection update

[SvenjaKern]

PR checklist:

• Tested by creator on localhost:8000/docs
• Tested by creator on refinery
• Tested by reviewer on localhost:8000/docs
• Tested by reviewer on refinery
• (If added) common code tested in notebook/ script
• Conforms with the agreed upon code pattern

[LeonardPuettmannKern]

Needs some work. :)

[LeonardPuettmannKern]

Would add this to the "llm" category as well as this is using a roberta model from HuggingFace.

[LeonardPuettmannKern]

This is supposed to be a classification brick and not an extraction one. This code tries to extract entities from a text and yield them, which is not how this model works. The output of the model returns a list containing a dict like this: [[{'label': 'disgust', 'score': 0.8062547445297241}, {'label': 'sadness', 'score': 0.107185959815979}, {'label': 'anger', 'score': 0.06451956927776337}, {'label': 'fear', 'score': 0.014504954218864441}, {'label': 'neutral', 'score': 0.0030063888989388943}, {'label': 'joy', 'score': 0.0026550842449069023}, {'label': 'surprise', 'score': 0.0018732782918959856}]]

Doing NER won't work.

[LeonardPuettmannKern]

@SvenjaKern Please watch out to never include API keys in your git commits. Please deactivate this API key and generate a new one. This is a public repository, which are often scanned for keys like this which can then be abused.

[LeonardPuettmannKern]

Name of the function call doesn't match up to the actual function.

[LeonardPuettmannKern]

Common code should be like this:

Suggested change

	nlp = spacy.load("en_core_web_sm")
	doc = nlp(text)
	for item in response_json:
	start = item["start"]
	end = item["end"]
	span = doc.char_span(start, end, alignment_mode="expand")
	ner_positions.append((item["entity_group"], span.start, span.end))
	return ner_positions
	except Exception as e:
	return f"That didn't work. Did you provide a valid API key? Go error: {e} and message: {response_json}"
	# ↑ necessary bricks function
	import requests
	def emotionality_detection(text, api_key):
	headers = {"Authorization": f"Bearer {api_key}"}
	data = {"inputs": text}
	try:
	response = requests.post("https://api-inference.huggingface.co/models/j-hartmann/emotion-english-distilroberta-base", headers=headers, json=data)
	response_json = response.json()
	# flatten the list of lists
	flat_list = [item for sublist in response_json for item in sublist]
	# find the item with the highest score
	max_item = max(flat_list, key=lambda x: x["score"])
	# retrieve the label of the item with the highest score
	max_label = max_item["label"]
	return max_label
	except Exception as e:
	return f"That didn't work. Did you provide a valid API key? Go error: {e} and message: {response_json}"
	# ↑ necessary bricks function
	# -----------------------------------------------------------------------------------------
	# ↓ example implementation
	def example_integration():
	hf_api_key = "<API_KEY_GOES_HERE>"
	texts = ["What a great day to go to the beach.", "Sorry to hear that. CAn I help you?", "Why the hell would you do that?"]
	for text in texts:
	output = emotionality_detection(text, api_key=hf_api_key)
	print(output)
	example_integration()