loggex 0.1.0

A flexible experiment logging library for research and machine learning projects

Loggex

Loggex helps you track your research experiments across multiple scripts by providing a central log file to record which script produced what results with which configuration.

Features

• Track experiment runs in CSV, JSON, or Excel format
• Load and add to your experiment log from any script
• Automatic timestamp tracking for each run
• Simple querying and filtering capabilities
• Quick summary reports of your experiments

Installation

pip install loggex

Quick Start

Creating a New Log

from loggex import Project

# Create a new experiment log (defaults to CSV format)
project = Project.new(
    name="sentiment_experiments",
    path="logs",
    columns=["experiment", "script", "model", "lr", "bs", "max_len", "f1_score", "results_file"]
)

# Log an experiment run
project.add(
    experiment="bert_base",
    script="train_bert.py",
    model="bert-base-uncased",
    lr=2e-5,
    bs=16,
    max_len=128,
    f1_score=0.91,
    results_file="results/bert_base_full.csv"
)

Adding Results from Another Script

from loggex import Project

# Load the existing log
project = Project.load("sentiment_experiments", path="logs")

# Add results from this script
project.add(
    experiment="bert_large",
    script="train_bert_improved.py",
    model="bert-large-uncased",
    lr=3e-5,
    bs=8,
    max_len=256,
    f1_score=0.94,
    results_file="results/bert_large_full.csv"
)

# View latest runs
project.tail()

Analyzing Your Experiments

# Generate a summary report
project.report()

# Filter experiments
best_runs = project.filter("f1_score > 0.92")
print(best_runs)

# Remove failed runs
project.remove("f1_score < 0.5")

# Export filtered results
project.export(
    format='excel',  # 'csv', 'json', or 'excel'
    query="model == 'bert-base-uncased'",
    path="analysis/bert_base_results.xlsx"
)

Using Different Formats

By default, Loggex uses CSV files, but you can choose other formats:

# Create a new log in Excel format
project = Project.new(
    name="mnist_experiments",
    path="logs",
    columns=["experiment", "script", "accuracy"],
    format='excel'  # 'csv', 'json', or 'excel'
)

Documentation

The Project class supports:

• Creating new logs (new())
• Loading existing logs (load())
• Adding log entries (add()) with automatic timestamp tracking
• Printing all logs (print())
• Viewing latest entries (tail(n=5))
• Removing entries (remove(query))
• Generating detailed reports (report())
• Exporting filtered results (export(format, query, path))

See in-line docstrings in the source code for complete usage details.

License

Licensed under the MIT License.