tracker-ml 0.0.6

Machine learning file and parameter version control SDK for www.tracker.ml

tracker_ml

www.tracker.ml SDK and CLI

Install

pip install tracker-ml

Build

Install build tools

python -m pip install --user --upgrade setuptools wheel

Create distribution archives

python setup.py sdist bdist_wheel

Install from archives

python -m pip install dist/tracker_ml-X.X.X.tar.gz

Python SDK

Use

First, initialize tracker using the CLI and update the files you want to track. Then import tracker_ml.tml anywhere and everywhere. Use is easy:

import tracker_ml.tml as tml

tml.login("username", "password")
tml.model("Logistic Regression")

# <machine learning code>

# record int, float, or str
tml.record("accuracy", 0.42)
tml.record("model", "Logistic Regression")

# record multiple values under the same key
tml.mrecord("epoch", 1)
tml.mrecord("epoch", 2)
tml.mrecord("epoch", 3)

# data will be saved locally and to the API on exit

All changes since the previous run and all recorded values will be automatically saved. The CLI can be used to view/compare trials and undo changes.

Command Line Interface

tracker.ml command line interface for locally tracking/reverting file changes and tracking results for each change. Similar to git, but works with the SDK to track every time a new model is trained/tested.

Use

Use the help command. (Not all commands displayed work yet)

$ tracker --help
$ tracker status --help

Initialize in the project root.

$ tracker init -u <username> -p <password> -n <project name>

Add file(s)/directory(s) that will be saved every run.

$ tracker add .

Stop recording file(s)/directory(s) that would be saved every run.

$ tracker remove .

View past trials and sort them

$ tracker status
 Total trials: 4
 Sorted by: id

  Id  |  Accuracy  |         Model
------------------------------------------
  4   |     63     |  Logistic Regression
  3   |     74     |  Logistic Regression
  2   |     50     |  Logistic Regression
  1   |     92     |  Logistic Regression
$ python tracker.py status -k accuracy -l 2 -r
 Total trials: 4
 Reverse sorted by: accuracy
 Only displaying 2 results

  Id  |  Accuracy  |         Model
------------------------------------------
  2   |     50     |  Logistic Regression
  4   |     63     |  Logistic Regression