Sync any experiment with Neptune

Create .json file containing your experiment data

You can run experiment in any language or framework and simply create a json with all the data that you care to save to neptune.

For example your experiment_data.json could look like this:

{
    'name':'example',
    'description':'json tracking experiment',
    'params':{'lr':0.1,
            'batch_size': 128,
            'dropount': 0.5
           },
    'properties':{'data_version':'1231ffwefef9',
                'data_path': '/mnt/path/to/data'
               },
    'tags':['resnet','no_preprocessing'],
    'upload_source_files':['train.sh'],
    'send_metric':{'log_loss': {'x': [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
                              'y': [0, 1, 4, 9, 16, 25, 36, 49, 64, 81, 100]
                               },
                    'accuracy':{'x': [0, 4, 5, 8, 9],
                                'y': [0.23, 0.47, 0.62, 0.89, 0.92]
                            }
                },
    'send_text':{ 'hash':{'x': [0, 4, 5, 8, 9],
                       'y': ['123123','as32e132', '123sdads', '123asdasd',' asd324132a']
                            },
               },
    'send_image':{'heatmaps':{'x': [0,1,2],
                                 'y': ['img1.png', 'img2.png', 'img3.png']
                                },
                 },
}

Sync with Neptune

Now all you need is to send this data to your Neptune project.

python -m neptunecontrib.sync.with_json \
--project_name neptune-ml/neptune-examples \
--filepath experiment_data.json

Explore your experiment in Neptune

Now your experiment data is safely logged to Neptune and you can easily explore it. Go to the experiment link to see for yourself!

image

image1

image2

[ ]: