​fastai is a deep learning library that provides practitioners with high-level components that can quickly and easily provide state-of-the-art results in standard deep learning domains and provides researchers with low-level components that can be mixed and matched to build new approaches.

What you will get with this integration?

With Neptune + fastai integration the following metadata is logged automatically for you:

• hyper-parameters

• Losses & metrics

• training code(Python scripts or Jupyter notebooks) and git information

• Dataset version

• Model Configuration, Architecture, and Weights

Installation

To install Neptune - fastai integration, go to your console and run:

pip install neptune-client[fastai]

Quickstart

Before you start

You need to have Python 3.6+ and install the following libraries installed:

• fastai

pip install fastai

 fastai==2.3.1

Step 1: Initialize a Neptune Run

Place this code snippet at the beginning of your script or notebook cell

import neptune.new as neptune
​
run = neptune.init(project = '<YOUR_WORKSPACE/YOUR_PROJECT>', 
                         api_token = '<YOURR_API_TOKEN>', 
                         source_files=['*.py'])

This opens a new Run in Neptune that allows you to log various objects.

Step 2: Add NeptuneCallback to learner or fit method:

To log metadata, you have to pass NeptuneCallback() to the callbacks argument of a learner or fit method.

Log a single training phase

learn = learner(...)
learn.fit(..., cbs = NeptuneCallback(run=run))

Log all training phases of the learner

learn = cnn_learner(..., cbs=NeptuneCallback(run=run))
learn.fit(...)
learn.fit(...)

Step 3: Stop Run

run.stop()

Step 4: Run your training script or notebook cell and monitor your training in Neptune UI

Run your script or notebook cell as you normally would.

After running your script or notebook cell you will get a link similar to: https://app.neptune.ai/o/common/org/fastai-integration/e/FAS-61 with common/fastai-integration replaced by your project, and FAS-61 replaced by your run.

Click on the link to open the Run in Neptune to watch your model training live.

Initially, it may be empty but keep the tab with the Run open to see your experiment metadata update in real-time.

More options

Customizing NeptuneCallback

You can change or extend the default behaviour of the NeptuneCallback() by passing callback a function to the constructor. For example:

def cb(self): 
    self.run['sys/name'] = 'Binary Classification'
    self.run['seed'] = 1000
            
neptune_cb = NeptuneCallback(before_fit=cb)

Log Model Architecture and Weights

• Add SaveModelCallback()

  You can log your model weight files during single training or all training phases add SavemodelCallback() to the callbacks' list of your learner or fit method.

• Log Every N epochs

n = 4
learn = learner(..., cbs=[
            SaveModelCallback(every_epoch=n),
            NeptuneCallback(run=run, upload_saved_models='all')])

• Best Model

​
learn = learner(..., cbs=[
           SaveModelCallback(), NeptuneCallback(run=run)])

Log Images and predictions

Plotting images and predictions is always helpful in computer vision tasks because it allows you to visually inspect the model's predictions.

With Neptune + fast.ai you can log torch tensors and they will be displayed as images in the Neptune UI.

from neptune.new.types import File
​
# Log image with predictions
img = torch.rand(30, 30, 3)
description = {"Predicted": pred, "Ground Truth": gt}
run['torch_tensor'].upload(File.as_image(img),
                                description = description)

How to ask for help?

Please visit the Getting help page. Everything regarding support is there.

Other pages you may like

You may want to check out other integrations from the Pytorch ecosystem: