Repères d'exemples

Travailler avec l'API d'OpenAI

James Chapman

Curriculum Manager, DataCamp

Fournir des exemples

Un robot examine des exemples

Repères d'exemples

Qu'est-ce que les repères d'exemples ?

Repères d'exemples : inclure des exemples pour guider les réponses de l'IA

Zero-shot : aucune exemple, seulement des consignes
One-shot : un exemple oriente la réponse
Few-shot : plusieurs exemples donnent plus de contexte

Représentation de zéro à quelques exemples

Pourquoi est-ce important ?

✅ Cas d'usage des repères d'exemples :

📑 Classification de texte
💬 Analyse de sentiments
🔍 Extraction de données
✨ … et bien d'autres !

Cas d'usage des repères d'exemples

Repère zéro exemple (Zero-shot)

prompt = """Classify sentiment as 1-5 (bad-good) in the following statements:
1. Meal was decent, but I've had better.
2. My food was delayed, but drinks were good.
"""

...

1. Meal was decent, but I've had better. - 3 (Neutral)
2. My food was delayed, but drinks were good. - 3 (Neutral)

Repère un exemple (One-shot)

prompt = """Classify sentiment as 1-5 (bad-good) in the following statements:
1. The service was very slow -> 1
2. Meal was decent, but I've had better. ->
3. My food was delayed, but drinks were good. ->
"""

1. The service was very slow -> 1
2. Meal was decent, but I've had better. -> 3
3. My food was delayed, but drinks were good. -> 4

Repères quelques exemples (Few-shot)

prompt = """Classify sentiment as 1-5 (bad-good) in the following statements:
1. The service was very slow -> 1
2. The steak was awfully good! -> 5
3. It was ok, no massive complaints. -> 3
4. Meal was decent, but I've had better. ->
5. My food was delayed, but drinks were good. ->
"""

1. The service was very slow -> 1
2. The steak was awfully good! -> 5
3. It was ok, no massive complaints. -> 3
4. Meal was decent, but I've had better. -> 4
5. My food was delayed, but drinks were good. -> 3

Catégorisation générale

prompt = """Classify the following animals as Land, Sea, or Both:
1. Blue whale
2. Polar bear
3. Salmon
4. Dog
"""

1. **Blue whale** - Sea
2. **Polar bear** - Both (primarily land but is a marine mammal that lives in sea ice)
3. **Salmon** - Both (spends part of its life in freshwater and part in the ocean)
4. **Dog** - Land

prompt = """Classify the following animals as Land, Sea, or Both:
1. Zebra = Land
2. Crocodile = Both
3. Blue whale = 
4. Polar bear = 
5. Salmon = 
6. Dog = 
"""

1. Zebra = Land
2. Crocodile = Both
3. Blue whale = Sea
4. Polar bear = Both
5. Salmon = Sea
6. Dog = Land

Passons à la pratique !

Travailler avec l'API d'OpenAI