Hello World! Meet Language AI

1
! pip install cohere altair umap-learn -q

1
import cohere
2
import pandas as pd
3
import numpy as np
4
import altair as alt
5
6
co = cohere.Client("COHERE_API_KEY") # Get your API key: https://dashboard.cohere.com/api-keys

1
prompt = "What is a Hello World program."
2
3
response = co.chat(
4
  message=prompt,
5
  model='command-r')
6
7
print(response.text)

1
prompt = """
2
Write the first paragraph of a blog post given a blog title.
3
--
4
Blog Title: Best Activities in Toronto
5
First Paragraph: Looking for fun things to do in Toronto? When it comes to exploring Canada's
6
largest city, there's an ever-evolving set of activities to choose from. Whether you're looking to
7
visit a local museum or sample the city's varied cuisine, there is plenty to fill any itinerary. In
8
this blog post, I'll share some of my favorite recommendations
9
--
10
Blog Title: Mastering Dynamic Programming
11
First Paragraph: In this piece, we'll help you understand the fundamentals of dynamic programming,
12
and when to apply this optimization technique. We'll break down bottom-up and top-down approaches to
13
solve dynamic programming problems.
14
--
15
Blog Title: Learning to Code with Hello, World!
16
First Paragraph:"""
17
18
response = co.chat(
19
  message=prompt,
20
  model='command-r')
21
22
print(response.text)

1
def generate_text(topic):
2
  prompt = f"""
3
Write the first paragraph of a blog post given a blog title.
4
--
5
Blog Title: Best Activities in Toronto
6
First Paragraph: Looking for fun things to do in Toronto? When it comes to exploring Canada's
7
largest city, there's an ever-evolving set of activities to choose from. Whether you're looking to
8
visit a local museum or sample the city's varied cuisine, there is plenty to fill any itinerary. In
9
this blog post, I'll share some of my favorite recommendations
10
--
11
Blog Title: Mastering Dynamic Programming
12
First Paragraph: In this piece, we'll help you understand the fundamentals of dynamic programming,
13
and when to apply this optimization technique. We'll break down bottom-up and top-down approaches to
14
solve dynamic programming problems.
15
--
16
Blog Title: {topic}
17
First Paragraph:"""
18
  # Generate text by calling the Chat endpoint
19
  response = co.chat(
20
    message=prompt,
21
    model='command-r')
22
23
  return response.text

1
topics = ["How to Grow in Your Career",
2
          "The Habits of Great Software Developers",
3
          "Ideas for a Relaxing Weekend"]

1
paragraphs = []
2
3
for topic in topics:
4
  paragraphs.append(generate_text(topic))
5
6
for topic,para in zip(topics,paragraphs):
7
  print(f"Topic: {topic}")
8
  print(f"First Paragraph: {para}")
9
  print("-"*10)

1
from cohere import ClassifyExample
2
3
examples = [
4
    ClassifyExample(text="I’m so proud of you", label="positive"),
5
    ClassifyExample(text="What a great time to be alive", label="positive"),
6
    ClassifyExample(text="That’s awesome work", label="positive"),
7
    ClassifyExample(text="The service was amazing", label="positive"),
8
    ClassifyExample(text="I love my family", label="positive"),
9
    ClassifyExample(text="They don't care about me", label="negative"),
10
    ClassifyExample(text="I hate this place", label="negative"),
11
    ClassifyExample(text="The most ridiculous thing I've ever heard", label="negative"),
12
    ClassifyExample(text="I am really frustrated", label="negative"),
13
    ClassifyExample(text="This is so unfair", label="negative"),
14
    ClassifyExample(text="This made me think", label="neutral"),
15
    ClassifyExample(text="The good old days", label="neutral"),
16
    ClassifyExample(text="What's the difference", label="neutral"),
17
    ClassifyExample(text="You can't ignore this", label="neutral"),
18
    ClassifyExample(text="That's how I see it", label="neutral")
19
]

1
inputs=["Hello, world! What a beautiful day",
2
        "It was a great time with great people",
3
        "Great place to work",
4
        "That was a wonderful evening",
5
        "Maybe this is why",
6
        "Let's start again",
7
        "That's how I see it",
8
        "These are all facts",
9
        "This is the worst thing",
10
        "I cannot stand this any longer",
11
        "This is really annoying",
12
        "I am just plain fed up"
13
        ]

1
def classify_text(inputs, examples):
2
  """
3
  Classify a list of input texts
4
  Arguments:
5
    inputs(list[str]): a list of input texts to be classified
6
    examples(list[Example]): a list of example texts and class labels
7
  Returns:
8
    classifications(list): each result contains the text, labels, and conf values
9
  """
10
  # Classify text by calling the Classify endpoint
11
  response = co.classify(
12
    model='embed-english-v2.0',
13
    inputs=inputs,
14
    examples=examples)
15
16
  classifications = response.classifications
17
18
  return classifications

1
predictions = classify_text(inputs,examples)
2
3
classes = ["positive","negative","neutral"]
4
for inp,pred in zip(inputs,predictions):
5
  class_pred = pred.predictions[0]
6
  class_idx = classes.index(class_pred)
7
  class_conf = pred.confidences[0]
8
9
  print(f"Input: {inp}")
10
  print(f"Prediction: {class_pred}")
11
  print(f"Confidence: {class_conf:.2f}")
12
  print("-"*10)

1
df = pd.read_csv("https://github.com/cohere-ai/notebooks/raw/main/notebooks/data/hello-world-kw.csv", names=["search_term"])
2
df.head()

1
def embed_text(texts, input_type):
2
  """
3
  Turns a piece of text into embeddings
4
  Arguments:
5
    text(str): the text to be turned into embeddings
6
  Returns:
7
    embedding(list): the embeddings
8
  """
9
  # Embed text by calling the Embed endpoint
10
  response = co.embed(
11
                model="embed-english-v3.0",
12
                input_type=input_type,
13
                texts=texts)
14
15
  return response.embeddings

1
df["search_term_embeds"] = embed_text(texts=df["search_term"].tolist(),
2
                                      input_type="search_document")
3
doc_embeds = np.array(df["search_term_embeds"].tolist())

1
query = "what is the history of hello world"
2
3
query_embeds = embed_text(texts=[query],
4
                          input_type="search_query")[0]

1
from sklearn.metrics.pairwise import cosine_similarity
2
3
def get_similarity(target, candidates):
4
  """
5
  Computes the similarity between a target text and a list of other texts
6
  Arguments:
7
    target(list[float]): the target text
8
    candidates(list[list[float]]): a list of other texts, or candidates
9
  Returns:
10
    sim(list[tuple]): candidate IDs and the similarity scores
11
  """
12
  # Turn list into array
13
  candidates = np.array(candidates)
14
  target = np.expand_dims(np.array(target),axis=0)
15
16
  # Calculate cosine similarity
17
  sim = cosine_similarity(target,candidates)
18
  sim = np.squeeze(sim).tolist()
19
20
  # Sort by descending order in similarity
21
  sim = list(enumerate(sim))
22
  sim = sorted(sim, key=lambda x:x[1], reverse=True)
23
24
  # Return similarity scores
25
  return sim

1
similarity = get_similarity(query_embeds,doc_embeds)
2
3
print("New query:")
4
print(new_query,'\n')
5
6
print("Similar queries:")
7
for idx,score in similarity[:5]:
8
  print(f"Similarity: {score:.2f};", df.iloc[idx]["search_term"])

1
import umap
2
reducer = umap.UMAP(n_neighbors=49)
3
umap_embeds = reducer.fit_transform(doc_embeds)
4
5
df['x'] = umap_embeds[:,0]
6
df['y'] = umap_embeds[:,1]

1
chart = alt.Chart(df).mark_circle(size=500).encode(
2
  x=
3
  alt.X('x',
4
      scale=alt.Scale(zero=False),
5
      axis=alt.Axis(labels=False, ticks=False, domain=False)
6
  ),
7
8
  y=
9
  alt.Y('y',
10
      scale=alt.Scale(zero=False),
11
      axis=alt.Axis(labels=False, ticks=False, domain=False)
12
  ),
13
14
  tooltip=['search_term']
15
  )
16
17
text = chart.mark_text(align='left', dx=15, size=12, color='black'
18
          ).encode(text='search_term', color= alt.value('black'))
19
20
result = (chart + text).configure(background="#FDF7F0"
21
      ).properties(
22
      width=1000,
23
      height=700,
24
      title="2D Embeddings"
25
      )
26
27
result