coinstac-sparse-dinunet 16.0.3

Distributed Sparse Neural Network implementation on COINSTAC (GPU).

coinstac-sparse-dinunet

Distributed Sparse Neural Network implementation on COINSTAC.

pip install coinstac-sparse-dinunet

Specify supported packages like pytorch & torchvision in a requirements.txt file

Highlights:

1. Creates sparse network based on single shot pruning SNIP algorithm (https://arxiv.org/abs/1810.02340). 
2. Performs distributed training and optimization with reduced bandwidth.
3. Automatic data splitting/k-fold cross validation.
4. Automatic model checkpointing.
5. GPU enabled local sites.
6. Customizable metrics(w/Auto serialization between nodes) to work with any schemes.
7. We can integrate any custom reduction and learning mechanism by extending coinstac_sparse_dinunet.distrib.reducer/learner.
...

---

Running an analysis in the coinstac App.

Add a new NN computation to COINSTAC (Development guide):

imports

from coinstac_sparse_dinunet import COINNDataset, COINNTrainer, COINNLocal
from coinstac_sparse_dinunet.metrics import COINNAverages, Prf1a

1. Define Data Loader

class MyDataset(COINNDataset):
    def __init__(self, **kw):
        super().__init__(**kw)
        self.labels = None

    def load_index(self, id, file):
        data_dir = self.path(id, 'data_dir') # data_dir comes from inputspecs.json
        ...
        self.indices.append([id, file])

    def __getitem__(self, ix):
        id, file = self.indices[ix]
        data_dir = self.path(id, 'data_dir') # data_dir comes from inputspecs.json
        label_dir = self.path(id, 'label_dir') # label_dir comes from inputspecs.json
        ...
        # Logic to load, transform single data item.
        ...
        return {'inputs':.., 'labels': ...}

2. Define Trainer

class MyTrainer(COINNTrainer):
    def __init__(self, **kw):
        super().__init__(**kw)

    def _init_nn_model(self):
        self.nn['model'] = MYModel(in_size=self.cache['input_size'], out_size=self.cache['num_class'])
    
    
   def single_iteration_for_masking(self, model, batch):
    
        #Defines sparsity level, loss function and other parameters to perform masking using SNIP
        
        sparsity_level = 0.85
        inputs, labels = batch['inputs'].to(self.device['gpu']).float(), batch['labels'].to(self.device['gpu']).long()
        indices = batch['ix'].to(self.device['gpu']).long()
        model.zero_grad()
        out = F.log_softmax(model.forward(inputs), 1)
        loss = F.nll_loss(out, labels)
        return {'out': out, 'loss': loss, 'indices': indices, 'sparsity_level': sparsity_level}



    def iteration(self, batch):
        inputs, labels = batch['inputs'].to(self.device['gpu']).float(), batch['labels'].to(self.device['gpu']).long()

        out = F.log_softmax(self.nn['model'](inputs), 1)
        loss = F.nll_loss(out, labels)
        _, predicted = torch.max(out, 1)
        score = self.new_metrics()
        score.add(predicted, labels)
        val = self.new_averages()
        val.add(loss.item(), len(inputs))
        return {'out': out, 'loss': loss, 'averages': val,
                'metrics': score, 'prediction': predicted}

---

Advanced use cases:

• Define custom metrics:

  • Extend coinstac_sparse_dinunet.metrics.COINNMetrics
  • Example: coinstac_sparse_dinunet.metrics.Prf1a for Precision, Recall, F1, and Accuracy

• Define Custom Learner / [custom Aggregator]/ custom Aggregator (Default is Distributed SGD)

Referenced from Trends Center coinstac-dinunet repository (https://github.com/trendscenter/coinstac-dinunet)