Approaches to modeling

yrodriguezmd · July 27, 2021, 12:08am

In Lesson 06_multicat, it initiated a code learn.model(x).
Screen Shot 2021-07-26 at 7.59.34 PM

I am not able to find a documentation for this ‘model’ term.

I was wondering why this step was needed, The previous lessons went straight to modeling from the DataBlock and dataloading steps. It did not pertain to a ‘model’ term.

Screen Shot 2021-07-26 at 7.58.40 PM

Is this because the variables are now in CPU?

Thank you!

Maria

meanpenguin · July 27, 2021, 2:19am

This step is not needed and is just an example on how to access the resnet18 model in the learner.

the .model is the actual resnet18 model that is embedded in the Learner.
So here, you are just manually getting some x values “dls.train.one_batch()” and feeding it to the resnet18 to get the output of the model (aka activations)

See here were the model is in the code

github.com

fastai/fastai/blob/cc7df3c17d2c25547772af0e6852bfb22bc0182a/fastai/learner.py#L92

    
      
          # Cell
          class Learner(GetAttr):
              _default='model'
              def __init__(self, dls, model, loss_func=None, opt_func=Adam, lr=defaults.lr, splitter=trainable_params, cbs=None,
                           metrics=None, path=None, model_dir='models', wd=None, wd_bn_bias=False, train_bn=True,
                           moms=(0.95,0.85,0.95)):
                  path = Path(path) if path is not None else getattr(dls, 'path', Path('.'))
                  if loss_func is None:
                      loss_func = getattr(dls.train_ds, 'loss_func', None)
                      assert loss_func is not None, "Could not infer loss function from the data, please pass a loss function."
                  self.dls,self.model = dls,model
                  store_attr(but='dls,model,cbs')
                  self.training,self.create_mbar,self.logger,self.opt,self.cbs = False,True,print,None,L()
                  self.add_cbs(L(defaults.callbacks)+L(cbs))
                  self("after_create")
          
          
    @property
              def metrics(self): return self._metrics
              @metrics.setter
              def metrics(self,v): self._metrics = L(v).map(mk_metric)

yrodriguezmd · July 27, 2021, 2:35am

I see it now, thank you!!