Quick Start¶

This guide will walk you through your first evaluation with TempoEval.

Prerequisites¶

Ensure you have installed the package:

pip install tempoeval

5-Minute "Hello World"¶

We will evaluate a simple scenario: * Query: "What happened in 2020?" * Doc A: "The COVID-19 pandemic began in 2020." (Relevant) * Doc B: "The 2008 financial crisis was severe." (Irrelevant)

Step 1: Import Components¶

from tempoeval.core import extract_qft, extract_dft
from tempoeval.metrics import TemporalPrecision

Step 2: Extract Focus Time¶

First, we need to understand what time the texts are about. We use extract_qft (Query Focus Time) and extract_dft (Document Focus Time).

query = "What happened in 2020?"
documents = [
    "The COVID-19 pandemic began in 2020.",
    "The 2008 financial crisis was severe."
]

# Extract
qft = extract_qft(query)
print(f"Query Time: {qft}") 
# Output: FocusTime(years={2020})

dfts = [extract_dft(doc) for doc in documents]
print(f"Doc Times: {dfts}") 
# Output: [FocusTime(years={2020}), FocusTime(years={2008})]

Step 3: Compute Metric¶

Now we verify if the retrieved documents match the query's time.

# Initialize Metric
metric = TemporalPrecision(use_focus_time=True)

# Compute
score = metric.compute(
    qft=qft, 
    dfts=dfts, 
    k=2  # Evaluated at top 2 documents
)

print(f"Precision@2: {score}")
# Output: 0.5 (Because 1 out of 2 docs was relevant)

What just happened?¶

Extraction: The library found "2020" in the query and the first document.
Intersection: It compared {2020} vs {2020} (Match!) and {2020} vs {2008} (No Match).
Scoring: It calculated Precision = (Matches / Total) = ½ = 0.5.

Next Steps¶

Now that you understand the basics, try evaluating a full RAG pipeline.

➡️ Evaluate Retrieval Tutorial