Prompt Strategies

Overview

In addition to calling the semantic operators, advanced prompt stratigies can be used to potentially get or improve the desired output. Lotus supports multiple reasoning strategies through the ReasoningStrategy enum.

Chain of Thought + Demonstrations:

Chain of Thought reasoning refers to structing prompts in a way that guides the model through a step-by-step process to arrive at a final answer. By breaking down complex tasks into intermediate steps, CoT ensures more accurate and logical output

Here is a simple example of using chain of thought with the Semantic Filter operator using COT

import pandas as pd

import lotus
from lotus.models import LM
from lotus.types import ReasoningStrategy

lm = LM(model="gpt-4o-mini")

lotus.settings.configure(lm=lm)
data = {
    "Course Name": [
        "Probability and Random Processes",
        "Optimization Methods in Engineering",
        "Digital Design and Integrated Circuits",
        "Computer Security",
    ]
}
df = pd.DataFrame(data)
user_instruction = "{Course Name} requires a lot of math"

example_data = {
    "Course Name": ["Machine Learning", "Reaction Mechanisms", "Nordic History"],
    "Answer": [True, True, False],
    "Reasoning": ["Machine Learning requires a solid understanding of linear alebra and calculus",
                  "Reaction Engineering requires Ordinary Differential Equations to solve reactor design problems",
                  "Nordic History has no math involved"]
}
examples = pd.DataFrame(example_data)

df = df.sem_filter(user_instruction, examples=examples, strategy=ReasoningStrategy.COT)
print(df)

When calling the Semantic Filter operator, we pass in an example DataFrame as well as the CoT strategy, which acts as a guide for how the model should reason and respond to the given instructions. For instance, in the examples DataFrame

• “Machine Learning” has an answer of True, with reasoning that it requires a solid understanding of linear algebra and calculus.

• “Reaction Mechanisms” also has an answer of True, justified by its reliance on ordinary differential equations for solving reactor design problems.

• “Nordic History” has an answer of False, as it does not involve any mathematical concepts.

Using the CoT strategy will provide an output below:

	Course Name		explanation_filter
0	Probability and Random Processes		Probability and Random Processes is heavily based on…
1	Optimization Methods in Engineering		Optimization Methods in Engineering typically involves…
2	Digital Design and Integrated Circuits | Digital Design and Integrated Circuits typically covers…

Here is another example that uses ZS_COT

import pandas as pd

import lotus
from lotus.models import LM
from lotus.types import ReasoningStrategy

lm = LM(model="ollama/deepseek-r1:7b", temperature=0.6)
lotus.settings.configure(lm=lm)

data = {
    "Review": [
        "This vacuum cleaner is the best I've ever owned. Highly recommend it!",
        "It's okay, not sure I would buy it again.",
        "Terrible experience, broke after a few uses.",
        "Amazing build quality and customer support. Would absolutely recommend.",
    ]
}
df = pd.DataFrame(data)

user_instruction = "{Review} suggests that the user would recommend the product to others"

df = df.sem_filter(
    user_instruction,
    strategy=ReasoningStrategy.ZS_COT,
    return_explanations=True,
    return_all=True
)

print(df)

Supported Reasoning Strategies

ReasoningStrategy.DEFAULT * Description: The default strategy. The model receives a plain instruction and is expected to provide a direct answer without any explicit reasoning.

ReasoningStrategy.COT (Chain of Thought) * Description: The model is prompted to reason step-by-step after being shown a few reasoning examples. Useful for tasks that benefit from intermediate steps to improve answer accuracy or interpretability.

• ReasoningStrategy.ZS_COT (Zero-Shot Chain of Thought)

Description: The model is instructed to reason step-by-step without examples. The reasoning process is triggered purely by the prompt (e.g., “Let’s think step by step”).

ReasoningStrategy.FEW_SHOT * Description: The model is provided with few-shot examples, but without explicit reasoning steps. It imitates the behavior from the examples to answer the final prompt.