scope 0.6.3

Metrics logging and analysis

🔬 Scope

Scalable metrics logging and analysis.

Features

• 🚀 Scalable: Quickly log and view petabytes of metrics, thousands of keys, and large videos.
• 🎞️ Formats: Log and view scalars, text, images, and videos. Easy to extend with custom formats.
• 🧑🏻‍🔬 Productivity: Metrics viewer with focus on power users with full keyboard support.
• ☁️ Cloud support: Directly write to and read from Cloud storage via pathlib interface.
• 🍃 Lightweight: The writer and reader measure only ~400 lines of Python code.
• 🧱 Reliable: Unit tested and used across diverse research projects.

Usage

Installation

pip install scope

Writing

import scope

writer = scope.Writer(logdir)

for step in range(3):
  writer.add(step, {
      'foo': 42,
      'bar': np.zeros((100, 640, 360, 3), np.uint8),
      'baz': 'Hello World',
  })
  writer.flush()

Viewing

python -m scope.viewer --basedir ... --port 8000

Reading

import scope

reader = scope.Reader(logdir)
print(reader.keys())               # ('foo', 'bar', 'baz')

print(reader.length('foo'))        # 3
steps, values = reader['foo']
print(steps)                       # np.array([0, 1, 2], np.int64)
print(values)                      # np.array([42, 42, 42], np.float64)

steps, filenames = reader['bar']
reader.load('bar', filenames[-1])  # np.zeros((100, 640, 360, 3), np.uint8)