### Adala Agent Quickstart Example

Source: https://github.com/humansignal/adala/blob/master/README.md

This snippet shows how to initialize an Adala agent with a static environment, a classification skill, and configure OpenAI chat runtimes for training and inference. It then trains the agent and runs predictions on a test dataset.

```python
import os
import pandas as pd

from adala.agents import Agent
from adala.environments import StaticEnvironment
from adala.skills import ClassificationSkill
from adala.runtimes import OpenAIChatRuntime
from rich import print

# Train dataset
train_df = pd.DataFrame([
    ["It was the negative first impressions, and then it started working.", "Positive"],
    ["Not loud enough and doesn't turn on like it should.", "Negative"],
    ["I don't know what to say.", "Neutral"],
    ["Manager was rude, but the most important that mic shows very flat frequency response.", "Positive"],
    ["The phone doesn't seem to accept anything except CBR mp3s.", "Negative"],
    ["I tried it before, I bought this device for my son.", "Neutral"],
], columns=["text", "sentiment"])

# Test dataset
test_df = pd.DataFrame([
    "All three broke within two months of use.",
    "The device worked for a long time, can't say anything bad.",
    "Just a random line of text."
], columns=["text"])

agent = Agent(
    # connect to a dataset
    environment=StaticEnvironment(df=train_df),

    # define a skill
    skills=ClassificationSkill(
        name='sentiment',
        instructions="Label text as positive, negative or neutral.",
        labels=["Positive", "Negative", "Neutral"],
        input_template="Text: {text}",
        output_template="Sentiment: {sentiment}"
    ),

    # define all the different runtimes your skills may use
    runtimes = {
        # You can specify your OpenRouter API Key here or set it ahead of time in your environment variable, OPENROUTER_API_KEY
        'openrouter': OpenAIChatRuntime(
            base_url="https://openrouter.ai/api/v1",
            model="anthropic/claude-3.5-haiku",
            api_key=os.getenv("OPENROUTER_API_KEY"),
            provider="Custom"
            ),
    },

    default_runtime='openrouter',

    teacher_runtimes = {
        "default" : OpenAIChatRuntime(
            base_url="https://openrouter.ai/api/v1",
            model="anthropic/claude-3.5-haiku",
            api_key=os.getenv("OPENROUTER_API_KEY"),
            provider="Custom"
        ),
    }
)

print(agent)
print(agent.skills)

agent.learn(learning_iterations=3, accuracy_threshold=0.95)

print('\n=> Run tests ...')
predictions = agent.run(test_df)
print('\n => Test results:')
print(predictions)

```

--------------------------------

### Adala Quickstart Example

Source: https://github.com/humansignal/adala/blob/master/docs/src/index.md

This Python code demonstrates how to set up and use an Adala agent for text classification. It includes defining training and testing datasets, configuring an agent with a classification skill and an OpenAI runtime, and then training and testing the agent.

```python
import pandas as pd

from adala.agents import Agent
from adala.environments import StaticEnvironment
from adala.skills import ClassificationSkill
from adala.runtimes import OpenAIChatRuntime
from rich import print

# Train dataset
train_df = pd.DataFrame([
    ["It was the negative first impressions, and then it started working.", "Positive"],
    ["Not loud enough and doesn't turn on like it should.", "Negative"],
    ["I don't know what to say.", "Neutral"],
    ["Manager was rude, but the most important that mic shows very flat frequency response.", "Positive"],
    ["The phone doesn't seem to accept anything except CBR mp3s.", "Negative"],
    ["I tried it before, I bought this device for my son.", "Neutral"],
], columns=["text", "sentiment"])

# Test dataset
test_df = pd.DataFrame([
    "All three broke within two months of use.",
    "The device worked for a long time, can't say anything bad.",
    "Just a random line of text."
], columns=["text"])

agent = Agent(
    # connect to a dataset
    environment=StaticEnvironment(df=train_df),

    # define a skill
    skills=ClassificationSkill(
        name='sentiment',
        instructions="Label text as positive, negative or neutral.",
        labels={'sentiment': ["Positive", "Negative", "Neutral"]},
        input_template="Text: {text}",
        output_template="Sentiment: {sentiment}"
    ),

    # define all the different runtimes your skills may use
    runtimes = {
        # You can specify your OPENAI API KEY here via `OpenAIRuntime(..., api_key='your-api-key')`
        'openai': OpenAIChatRuntime(model='gpt-3.5-turbo'),
    },
    default_runtime='openai',
    
    # NOTE! If you have access to GPT-4, you can uncomment the lines bellow for better results
#     default_teacher_runtime='openai-gpt4',
#     teacher_runtimes = {
#       'openai-gpt4': OpenAIRuntime(model='gpt-4')
#     }
)

print(agent)
print(agent.skills)

agent.learn(learning_iterations=3, accuracy_threshold=0.95)

print('\n=> Run tests ...')
predictions = agent.run(test_df)
print('\n => Test results:')
print(predictions)

```

--------------------------------

### Adala Quickstart with Pandas and OpenAI

Source: https://github.com/humansignal/adala/blob/master/README.md

Demonstrates setting up an Adala agent for text classification using pandas DataFrames and the OpenAI GPT-4o model. This example includes data preparation, agent configuration with a classification skill, and running the agent for learning and prediction.

```python
import pandas as pd

from adala.agents import Agent
from adala.environments import StaticEnvironment
from adala.skills import ClassificationSkill
from adala.runtimes import OpenAIChatRuntime
from rich import print

# Train dataset
train_df = pd.DataFrame([
    ["It was the negative first impressions, and then it started working.", "Positive"],
    ["Not loud enough and doesn't turn on like it should.", "Negative"],
    ["I don't know what to say.", "Neutral"],
    ["Manager was rude, but the most important that mic shows very flat frequency response.", "Positive"],
    ["The phone doesn't seem to accept anything except CBR mp3s.", "Negative"],
    ["I tried it before, I bought this device for my son.", "Neutral"],
], columns=["text", "sentiment"])

# Test dataset
test_df = pd.DataFrame([
    "All three broke within two months of use.",
    "The device worked for a long time, can't say anything bad.",
    "Just a random line of text."
], columns=["text"])

agent = Agent(
    # connect to a dataset
    environment=StaticEnvironment(df=train_df),

    # define a skill
    skills=ClassificationSkill(
        name='sentiment',
        instructions="Label text as positive, negative or neutral.",
        labels=["Positive", "Negative", "Neutral"],
        input_template="Text: {text}",
        output_template="Sentiment: {sentiment}"
    ),

    # define all the different runtimes your skills may use
    runtimes = {
        # You can specify your OPENAI API KEY here via `OpenAIRuntime(..., api_key='your-api-key')`
        'openai': OpenAIChatRuntime(model='gpt-4o'),
    },
    teacher_runtuntimes = {
        # You can specify your OPENAI API KEY here via `OpenAIRuntime(..., api_key='your-api-key')`
        'default': OpenAIChatRuntime(model='gpt-4o'),
    },
    default_runtime='openai',
)

print(agent)
print(agent.skills)

agent.learn(learning_iterations=3, accuracy_threshold=0.95)

print('\n=> Run tests ...')
predictions = agent.run(test_df)
print('\n => Test results:')
print(predictions)

```

--------------------------------

### Install Adala from GitHub

Source: https://github.com/humansignal/adala/blob/master/README.md

Install the most up-to-date version of Adala directly from its GitHub repository.

```sh
pip install git+https://github.com/HumanSignal/Adala.git
```

--------------------------------

### Install Project Dependencies

Source: https://github.com/humansignal/adala/blob/master/docs/README.md

Installs the required Python packages for documentation development using pip.

```bash
pip install requirements.txt
```

--------------------------------

### System Prompt Example

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

An example of a system prompt for a helpful assistant. This sets the initial behavior for the language model.

```text
Result:\n"{system}":\nYou are a helpful assistant.
```

--------------------------------

### Start Kafka and Redis (Native)

Source: https://github.com/humansignal/adala/blob/master/server/README.md

Use this command to start Kafka and Redis when running the application natively. Ensure you are in the repository root directory.

```bash
cd ..
docker-compose -f docker-compose.native.yml up
```

--------------------------------

### Install Development Dependencies with Poetry

Source: https://github.com/humansignal/adala/blob/master/CONTRIBUTION.md

Use this command to install all necessary dependencies for development, including optional ones.

```bash
poetry install --with dev
```

--------------------------------

### Serve MkDocs Documentation Locally

Source: https://github.com/humansignal/adala/blob/master/docs/README.md

Starts a local development server with live reloading for MkDocs documentation.

```bash
mkdocs serve
```

--------------------------------

### Start Adala Client (Development)

Source: https://github.com/humansignal/adala/blob/master/server/README.md

Command to start the client application for development convenience. This is not used in production.

```bash
cd ui/
yarn start
```

--------------------------------

### Example 0: Pizza Pieces Calculation

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example demonstrates a correct prediction for a word problem involving percentages and fractions of pizzas. It details the step-by-step calculation process.

```text
Q: Ann, Bill, Cate, and Dale each buy personal pan pizzas cut into 4 pieces. If Bill and Dale eat 50% of their 
pizzas and Ann and Cate eat 75% of the pizzas, how many pizza pieces are left uneaten?
A: 
To solve this problem, we need to first determine the total number of pizza pieces that were bought. Since each 
person bought a personal pan pizza cut into 4 pieces, we can calculate the total number of pieces by multiplying 4 
by the number of people, which is 4. This gives us a total of 16 pizza pieces.

Next, we need to calculate how many pieces were eaten by each person. Bill and Dale each ate 50% of their pizzas, 
which is equivalent to 2 pieces each (50% of 4 pieces is 2 pieces). Ann and Cate each ate 75% of their pizzas, 
which is equivalent to 3 pieces each (75% of 4 pieces is 3 pieces).

To find the total number of pieces eaten, we can add up the pieces eaten by each person: 2 + 2 + 3 + 3 = 10 pieces.

Finally, to find the number of pieces left uneaten, we can subtract the total number of pieces eaten from the total
number of pieces bought: 16 - 10 = 6 pieces.

Therefore, there are 6 pizza pieces left uneaten.
```

--------------------------------

### Developer Installation of Adala

Source: https://github.com/humansignal/adala/blob/master/README.md

Set up Adala for development by cloning the repository and installing dependencies using poetry.

```sh
git clone https://github.com/HumanSignal/Adala.git
cd Adala/
poetry install
```

--------------------------------

### Example 3: Objective Input

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

An example of an objective product review and its correct classification.

```text
Input: The phone doesn't seem to accept anything except CBR mp3s

Output: Objective

User feedback: Prediction is correct.
```

--------------------------------

### Example 0: Subjective Input

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

An example of a subjective product review and its correct classification.

```text
Input: The mic is great.

Output: Subjective

User feedback: Prediction is correct.
```

--------------------------------

### Start Adala Server

Source: https://github.com/humansignal/adala/blob/master/server/worker_pool/README.md

Starts the Adala server for API interactions. Ensure this is running before submitting data.

```bash
python -m server.app
```

--------------------------------

### Serve Documentation Locally

Source: https://github.com/humansignal/adala/blob/master/CONTRIBUTION.md

Build and serve the project's documentation locally using mkdocs. Ensure development dependencies are installed via poetry.

```bash
mkdocs serve -f ./docs/mkdocs.yml
```

--------------------------------

### GSM8k Example 3: Initial Money Calculation

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example involves working backward to find an initial amount of money after several spending percentages and a fixed amount are deducted. User feedback indicates the prediction was correct.

```text
Q: Abigail spent 60% of her money on food, and 25% of the remainder on her phone bill. After spending $20 on 
entertainment, she is left with $40. How much money did Abigail have initially?

A: 
Let x be the initial amount of money Abigail had.

After spending 60% on food, she is left with 40% of x, which is 0.4x.

Out of this 0.4x, she spends 25% on her phone bill, which is 0.25(0.4x) = 0.1x.

After spending $20 on entertainment, she is left with 0.4x - 0.1x - $20 = $40.

Simplifying, we get 0.3x = $60.

Dividing both sides by 0.3, we get x = $200.

Therefore, Abigail initially had $200.

User feedback: Prediction is correct.
```

--------------------------------

### Install Label Studio

Source: https://github.com/humansignal/adala/blob/master/README.md

Install Label Studio, a recommended tool for human-in-the-loop labeling and ground truth dataset production, which Adala supports.

```sh
pip install label-studio
```

--------------------------------

### Learning Agent Prompt and Examples

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

This snippet shows the initial prompt for the Learning Agent, designed to classify statements as subjective or objective, along with several examples demonstrating its usage and expected outputs. It also includes user feedback on prediction accuracy.

```text
Result:
"{user}":

## Current prompt
"Analyze the provided product review and ascertain whether it predominantly expresses 'Subjective' or 'Objective' 
statements. Classify it as 'Subjective' if the review primarily discloses personal feelings, emotions, or 
preferences. However, should the review describe factual details about the product’s functionality or performance, 
even if these details stem from personal experiences, yet do not assert personal emotions or preferences, classify 
it as 'Objective'. Keep in mind that a review can draw from personal experiences yet still remain 'Objective' if it
imparts information about the product's performance or functionality without declaring personal emotions or 
preferences."

## Examples
### Example #0

Input: The mic is great.

Output: Subjective

User feedback: Prediction is correct.


### Example #1

Input: Will order from them again!

Output: Subjective

User feedback: Prediction is correct.


### Example #2

Input: Not loud enough and doesn't turn on like it should.

Output: Subjective

User feedback: Prediction is incorrect. Correct answer: "Objective"


### Example #3

Input: The phone doesn't seem to accept anything except CBR mp3s

Output: Objective

User feedback: Prediction is correct.


### Example #4

Input: All three broke within two months of use.

Output: Objective

User feedback: Prediction is correct.



Summarize your analysis about incorrect predictions and suggest changes to the prompt.

```

--------------------------------

### Example 1: Subjective Input

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Another example of a subjective product review and its correct classification.

```text
Input: Will order from them again!

Output: Subjective

User feedback: Prediction is correct.
```

--------------------------------

### Example 1: Paintball Cost Calculation

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example shows a correct prediction for a word problem involving multiplication to calculate total monthly expenses. It breaks down the calculation into clear steps.

```text
Q: John plays paintball 3 times a month.  Each time he plays he buys 3 boxes of paintballs.  They cost $25 per box.
How much does he spend a month on paintballs?
A: 
To determine how much John spends on paintballs in a month, we need to multiply the number of times he plays by the
number of boxes he buys each time, and then multiply that by the cost per box.

Step 1: Determine the number of boxes John buys in a month.
Since John plays paintball 3 times a month and buys 3 boxes each time, we can multiply 3 by 3 to get the total 
number of boxes he buys in a month: 3 x 3 = 9 boxes.

Step 2: Determine the cost of 9 boxes of paintballs.
Each box costs $25, so we can multiply 9 by $25 to get the total cost of 9 boxes: 9 x $25 = $225.
```

--------------------------------

### Install Adala Package

Source: https://github.com/humansignal/adala/blob/master/docs/src/index.md

Install the Adala library using pip. This command should be run in your terminal or command prompt.

```sh
pip install adala
```

--------------------------------

### Example 4: Objective Input

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Another example of an objective product review and its correct classification.

```text
Input: All three broke within two months of use.

Output: Objective

User feedback: Prediction is correct.
```

--------------------------------

### Start ADALA Server

Source: https://github.com/humansignal/adala/blob/master/README.md

Starts the ADALA server on a specified port. Use the default port if none is provided.

```sh
adala start --port 8090
```

--------------------------------

### GSM8k Example 0: Bonus Calculation

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example demonstrates a word problem involving calculating hourly wage with a performance bonus, including extra hours. User feedback indicates the prediction was correct.

```text
Q: John works a job that offers performance bonuses.  He makes $80 a day and works for 8 hours.  He has the option 
of working hard to earn the performance bonus of an extra $20 a day, but the extra effort results in a 2-hour 
longer workday.  How much does John make per hour if he decides to earn the bonus?

A: John makes $10 per hour if he decides to earn the bonus.

User feedback: Prediction is correct.
```

--------------------------------

### System Prompt Template

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Defines the system's role and initial instructions for the language model. This is the starting point for assistant interactions.

```text
Result:\n"{system}":\nYou are a helpful assistant.\n\n
```

--------------------------------

### Skill Evaluation Example 1

Source: https://github.com/humansignal/adala/blob/master/examples/skillsets_sequence_of_skills.ipynb

Shows the output for 'skill_0' with 100.00% accuracy.

```text
Skill output to improve:  "skill_0"  ( Skill = "skill_0" )
Accuracy =  100.00 %
```

--------------------------------

### Prompt Engineering Workflow - Step 1

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

Describes the first step in a prompt engineering process where the user provides the current prompt and examples for analysis.

```text
Result:
"{user}":

## Step 1
I will provide you with the current prompt along with prediction examples. Each example contains the input text, 
the final prediction produced by the model, and the user feedback. User feedback indicates whether the model 
prediction is correct or not. Your task is to analyze the examples and user feedback, determining whether the 
existing instruction is describing the task reflected by these examples precisely, and suggests changes to the 
prompt to address the incorrect predictions.
```

--------------------------------

### Skill Evaluation Example 2

Source: https://github.com/humansignal/adala/blob/master/examples/skillsets_sequence_of_skills.ipynb

Shows the output for 'skill_1' with 0.00% accuracy, indicating a need for improvement.

```text
Skill output to improve:  "skill_1"  ( Skill = "skill_1" )
Accuracy =  0.00 %
```

--------------------------------

### Start Adala Application (Native)

Source: https://github.com/humansignal/adala/blob/master/server/README.md

Command to start the Uvicorn server for the Adala application when running natively. It binds to all network interfaces on port 30001.

```bash
poetry run uvicorn app:app --host 0.0.0.0 --port 30001
```

--------------------------------

### Progress Bar Example

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

Shows a progress bar indicating task completion, likely during a batch processing or iterative task.

```text
Output:
100%|█| 5/5 [00:15<00:00, 

```

--------------------------------

### Progress Output Example

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Displays the progress of a task, likely related to data processing or model training, showing completion percentage and rate.

```text
Output:
100%|█| 5/5 [00:01<00:00, 
```

--------------------------------

### Classification Example: Furniture/Home Decor (Correct)

Source: https://github.com/humansignal/adala/blob/master/examples/classification_skill.ipynb

An example of correct classification for a product description falling under the 'Furniture/Home Decor' category.

```text
Text: Wooden cream for surfaces.

Category: Furniture/Home Decor
```

--------------------------------

### Start Skill Learning Process

Source: https://github.com/humansignal/adala/blob/master/README.md

Initiates the learning process for a defined skill. ADALA will use the uploaded data to train the skill.

```sh
adala learn --skill "Object Classification"
```

--------------------------------

### Set OpenAI API Key

Source: https://github.com/humansignal/adala/blob/master/README.md

Configure your environment to use OpenAI services by exporting your API key. Refer to OpenAI's documentation for key setup instructions.

```sh
export OPENAI_API_KEY='your-openai-api-key'
```

--------------------------------

### GSM8k Example 2: Rattlesnake Calculation

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example requires calculating the number of rattlesnakes in a park given the total number of snakes, the count of boa constrictors, and the ratio of pythons to boa constrictors. User feedback indicates the prediction was correct.

```text
Q: There are 200 snakes in a park. There are three times as many pythons as boa constrictors. If there 40 boa 
constrictors and the rest of the snakes are rattlesnakes, calculate the total number of rattlesnakes in the park.

A: 200 - 40 = 160
160/4 = 40
There are 40 rattlesnakes in the park.

User feedback: Prediction is correct.
```

--------------------------------

### Prepare Test Data

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Create a Pandas DataFrame with text data to be used for testing the agent's skill. This step is crucial for feeding real-world examples into the model.

```python
test_df = pd.DataFrame([
    "Doesn't hold charge.",
    "Excellent bluetooth headset",
    "I love this thing!",
    "VERY DISAPPOINTED."
], columns=['text'])
test_df
```

--------------------------------

### Add a New Runtime

Source: https://github.com/humansignal/adala/blob/master/CONTRIBUTION.md

Introduce a new runtime by following the structure of the Runtime abstract class. This example provides a basic structure.

```python


```

--------------------------------

### Start Adala Server

Source: https://github.com/humansignal/adala/blob/master/README.md

Initiates the Adala server on the default port 8090. Each agent runs as its own web server.

```sh
# Start the Adala server on default port 8090
adala start
```

--------------------------------

### Assistant's Readiness to Help

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Confirms the assistant's willingness to help with prompt engineering tasks. Indicates readiness to receive prompt and examples.

```text
Result:\n"{assistant}":\nSure, I’d be happy to help you with this prompt engineering problem. Please provide me with the current prompt and \nthe examples with user feedback.\n\n
```

--------------------------------

### Start Celery Workers (Native)

Source: https://github.com/humansignal/adala/blob/master/server/README.md

Initiates Celery workers for stream inference. Navigate to the tasks directory before running this command.

```bash
cd tasks/
poetry run celery -A stream_inference worker --loglevel=info
```

--------------------------------

### Example 2: Celine's Shopping Trip

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

Illustrates the initial step of identifying conditions for a word problem involving purchases and calculating change.

```text
### Example # 1

Q: A shop sells laptops at $ 600 each and a smartphone at $ 400.  Celine buys two laptops and four smartphones for 
her children. How much change does she get back if she has $ 3000? 

A: 
Step  1: Identify and understand the relationships and conditions described in the problem.
- Celine is buying two laptops and four smartphones for her children.
- The laptops cost $ 600 each and the smartphones cost $ 400 each.
- Celine has $ 3000 to spend.


```

--------------------------------

### Analysis of Incorrect Predictions and Prompt Suggestions

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This section summarizes the analysis of incorrect predictions and suggests changes to the prompt to improve accuracy. It highlights the need for clearer instructions or examples to guide the model's reasoning process.

```text
Summarize your analysis about incorrect predictions and suggest changes to the prompt.
```

--------------------------------

### Build MkDocs Documentation

Source: https://github.com/humansignal/adala/blob/master/docs/README.md

Builds the static HTML documentation site for deployment.

```bash
mkdocs build
```

--------------------------------

### Classification Example: Footwear/Clothing

Source: https://github.com/humansignal/adala/blob/master/examples/classification_skill.ipynb

An example of correct classification for a product description falling under the 'Footwear/Clothing' category.

```text
Text: Chocolate leather boots.

Category: Footwear/Clothing
```

--------------------------------

### Classification Example: Electronics

Source: https://github.com/humansignal/adala/blob/master/examples/classification_skill.ipynb

An example of correct classification for a product description falling under the 'Electronics' category.

```text
Text: Apple product with a sleek design.

Category: Electronics
```

--------------------------------

### Initialize Adala Client and Submit Batch

Source: https://github.com/humansignal/adala/blob/master/server/worker_pool/README.md

Demonstrates how to initialize the AdalaClient and submit a batch of data for processing using the client.

```python
from label_studio_enterprise.lse_ml_models.adala_client import AdalaClient

# Initialize client
client = AdalaClient()

# Submit batch for processing
result = client.submit_worker_pool_batch(model_run, data)
```

--------------------------------

### Initialize and Run Classification Agent

Source: https://github.com/humansignal/adala/blob/master/examples/ontology_creator.ipynb

Sets up a classification agent using ClassificationSkill and OpenAI's GPT-4 model, then runs predictions on a test dataset and visualizes the results.

```python
from adala.skills import ClassificationSkill

predictor = Agent(
    skills=LinearSkillSet(skills=[ClassificationSkill(labels={'label': labels})]),
    runtimes={
        'default': OpenAIChatRuntime(model='gpt-4-1106-preview', verbose=False)
    })

df_test = pd.DataFrame(data=ds['test'][:100]['text'], columns=['text'])

pred_test = predictor.run(df_test)
pred_test.label.value_counts().plot.pie(figsize=(12, 12))
```

--------------------------------

### Create New MkDocs Project

Source: https://github.com/humansignal/adala/blob/master/docs/README.md

Command to create a new MkDocs project directory.

```bash
mkdocs new [dir-name]
```

--------------------------------

### Classification Example: Beauty/Personal Care

Source: https://github.com/humansignal/adala/blob/master/examples/classification_skill.ipynb

An example of correct classification for a product description falling under the 'Beauty/Personal Care' category.

```text
Text: Natural finish for your lips.

Category: Beauty/Personal Care
```

--------------------------------

### Start Celery Workers

Source: https://github.com/humansignal/adala/blob/master/server/worker_pool/README.md

Starts Celery workers to process tasks from the 'default' queue. Monitor logs for activity.

```bash
celery -A server.celery_app worker --loglevel=info --queues=default
```

--------------------------------

### Displaying Agent's Learned Instructions

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This Python snippet prints the current instructions for the agent's math_solver skill, showing the refined prompt after learning.

```python
print(agent.skills['math_solver'].instructions)
```

--------------------------------

### Incorrect Prediction Example

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This snippet shows an example of an incorrect prediction from a math reasoning model, highlighting the correct answer and the steps to reach it.

```text
Prediction is incorrect. Correct   answer: "In 7 years, Kaylee will be 5 * 3 = <<5*3=15>>15 years old Let x be the age Kaylee is now. After 7 years, Kaylle will be 15 years old, so x + 7 = 15 x = 15 - 7 x = <<8=8>>8 years old #### 8"
```

--------------------------------

### Example 2: Subjective Input with Incorrect Prediction

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

An example where the model incorrectly predicted 'Subjective' for a review that should be 'Objective'. This highlights a potential ambiguity in the prompt.

```text
Input: Not loud enough and doesn't turn on like it should.

Output: Subjective

User feedback: Prediction is incorrect. Correct answer:  "Objective" 
```

--------------------------------

### Classification Example: Furniture/Home Decor (Incorrect)

Source: https://github.com/humansignal/adala/blob/master/examples/classification_skill.ipynb

An example where the initial prompt led to an incorrect classification. The model predicted 'Electronics' instead of 'Furniture/Home Decor'.

```text
Text: Laptop stand for the kitchen.

Category: Electronics
```

--------------------------------

### Initialize and Run Summarization Skill

Source: https://github.com/humansignal/adala/blob/master/examples/summarization_skill.ipynb

Sets up an Adala Agent with the SummarizationSkill and runs it on the prepared DataFrame. The skill processes the 'text' column to generate summaries.

```python
from adala.agents import Agent
from adala.skills.collection.summarization import SummarizationSkill

agent = Agent(
    skills=SummarizationSkill(
        name='summarization',
        input_data_field='text'
    )
)

agent.run(df)
```

--------------------------------

### GSM8k Example 4: Hour Difference Calculation

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example calculates the difference in hours worked between two pairs of weeks. User feedback indicates the prediction was incorrect, and a detailed breakdown of the correct calculation is provided.

```text
Q: Davida worked 35 hours on each of Weeks 1 and 2. She worked 48 hours each of Weeks 3 and 4. How many more hours 
did Davida work on Weeks 3 and 4 than on Weeks 1 and 2?

A: 16 hours

User feedback: Prediction is incorrect. Correct answer: "Week 1 + Week 2 = 35 + 35 = 70 hours
Week 3 + Week 4 = 48 + 48 = 96 hours
Subtract Weeks 3-4 from Weeks 1-2 = 96 - 70 = 26
#### 26"
```

--------------------------------

### GSM8k Example 1: Annual Insurance Cost

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This example involves calculating the annual cost of car insurance based on a monthly payment and a percentage of the total cost. User feedback indicates the prediction was correct.

```text
Q: Nancy agreed to pay 40% of the cost of her daughter's car insurance, which costs $80 a month. How much will 
Nancy pay each year?

A: Nancy will pay $384 each year for her daughter's car insurance.

User feedback: Prediction is correct.
```

--------------------------------

### User Feedback and Prompt Refinement Guidance

Source: https://github.com/humansignal/adala/blob/master/examples/gsm8k_test.ipynb

This snippet outlines the guidance for refining a prompt based on user feedback, focusing on precise task description and addressing identified issues.

```text
Now please carefully review your reasoning in Step  1: Getting feedback, analyzing and improving  ... 

## Current prompt


## Follow this guidance to refine the prompt:

1. The new prompt should should describe the task precisely, and address the points raised in the user feedback.    
2. The new prompt should be similar to the current instruction, and only differ in the parts that address the 
issues you identified in Step  1.    
Example:
    - Current prompt:  "The model should generate a summary of the input text." 
    - New prompt:  "The model should generate a summary of the input text. Pay attention to the original style."        
3. Reply only with the new prompt. Do not include input and output templates in the prompt.
```

--------------------------------

### User Prompt and Template Structure

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

Illustrates how user input and prompts are structured and concatenated to form the final input for the LLM. Shows the template for prompt, input text, and output prediction.

```text
Result:\n"{user}":\n\nA prompt is a text paragraph that outlines the expected actions and instructs the large language model (LLM) to \ngenerate a specific output. This prompt is concatenated with the input text, and the model then creates the \nrequired output.\nThis describes the full template how the prompt is concatenated with the input to produce the output:\n\n```\n\n{prompt}\nInput: {text}\nOutput: {prediction}\n```\n\nHere:\n- "Input: {text}" is input template,\n- "{prompt}" is the LLM prompt,\n- "Output: {prediction}" is the output template.\n\nModel can produce erroneous output if a prompt is not well defined. In our collaboration, we’ll work together to \nrefine a prompt. The process consists of two main steps:\n\n## Step 1\nI will provide you with the current prompt along with prediction examples. Each example contains the input text, \nthe final prediction produced by the model, and the user feedback. User feedback indicates whether the model \nprediction is correct or not. Your task is to analyze the examples and user feedback, determining whether the \nexisting instruction is describing the task reflected by these examples precisely, and suggests changes to the \nprompt to address the incorrect predictions.\n\n## Step 2\nNext, you will carefully review your reasoning in step 1, integrate the insights to refine the prompt, and provide \nme with the new prompt that improves the model’s performance.\n\n
```

--------------------------------

### Display MkDocs Help Message

Source: https://github.com/humansignal/adala/blob/master/docs/README.md

Prints the help message for MkDocs commands and exits.

```bash
mkdocs -h
```

--------------------------------

### Assistant Response to Prompt Engineering Request

Source: https://github.com/humansignal/adala/blob/master/examples/quickstart.ipynb

The assistant's confirmation to help with prompt engineering, requesting the current prompt and examples.

```text
Result:\n"{assistant}":\nSure, I’d be happy to help you with this prompt engineering problem. Please provide me with the current prompt and \nthe examples with user feedback.\n
```