https://github.com/takuseno/nnabla-mlflow

mlflow utilities for nnabla
https://github.com/takuseno/nnabla-mlflow

nnabla

Last synced: about 2 months ago
JSON representation

mlflow utilities for nnabla

• Host: GitHub
• URL: https://github.com/takuseno/nnabla-mlflow
• Owner: takuseno
• Created: 2020-02-18T06:51:03.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2020-02-27T08:40:34.000Z (almost 5 years ago)
• Last Synced: 2024-10-23T04:06:26.112Z (3 months ago)
• Topics: nnabla
• Language: Python
• Size: 2.93 KB
• Stars: 3
• Watchers: 3
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# nnabla-mlflow

MLflow utilities for NNabla.

## dependencies

- nnabla

- mlflow

## install

```

$ pip install git+https://github.com/takuseno/nnabla-mlflow

```

## usage

Calling `nnabla_mlflow.autolog` monkey-pathches NNabla's monitors to automatically save data as MLflow metrics.

```py

import mlflow

import nnabla_mlflow

from nnabla.monitor import MonitorSeries

nnabla_mlflow.autolog() # monkey-patch Monitor classes

metric_monitor = MonitorSeries('metric', interval=1)

with mlflow.start_run():

    metric_monitor.add(1, 2) # mlflow.log_metric('metric', 1, step=2) is internally called

```

Currently, the following monitors are supported.

- MonitorSeries

- MonitorTimeElapsed

- MonitorImage

- MonitorImageTile

To automatically save parameters as an artifact, pass `with_save_parameters=True`.

```py

nnabla_mlflow.autolog(with_save_parameters=True)

```

## TODO

- [x] support MonitorImage and MonitorImageTile

- [ ] support `log_model` and `save_model` just like `mlflow.tensorflow` and `mlflow.pytorch`

- [ ] support NNabla's Trainer