Importing Previous Models

[Albertotono]

I exported successully one of my model trained with

Resnet34 for 4 Categories images Resize to 224

Now, I would like to start the training of the Multicategory from my previous model, mentioned above.

I loaded in this way and then
learn = load_learner(path/'export.pkl')

I ran this

x,y = dls.train.one_batch()
activs = learn.model(x)
activs.shape

And I got

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

RuntimeError                              Traceback (most recent call last)
<ipython-input-19-7b26555ca519> in <module>
      1 x,y = dls.train.one_batch()
----> 2 activs = learn.model(x)
      3 activs.shape

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    530             result = self._slow_forward(*input, **kwargs)
    531         else:
--> 532             result = self.forward(*input, **kwargs)
    533         for hook in self._forward_hooks.values():
    534             hook_result = hook(self, input, result)

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/container.py in forward(self, input)
     98     def forward(self, input):
     99         for module in self:
--> 100             input = module(input)
    101         return input
    102 

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    530             result = self._slow_forward(*input, **kwargs)
    531         else:
--> 532             result = self.forward(*input, **kwargs)
    533         for hook in self._forward_hooks.values():
    534             hook_result = hook(self, input, result)

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/container.py in forward(self, input)
     98     def forward(self, input):
     99         for module in self:
--> 100             input = module(input)
    101         return input
    102 

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    530             result = self._slow_forward(*input, **kwargs)
    531         else:
--> 532             result = self.forward(*input, **kwargs)
    533         for hook in self._forward_hooks.values():
    534             hook_result = hook(self, input, result)

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/conv.py in forward(self, input)
    343 
    344     def forward(self, input):
--> 345         return self.conv2d_forward(input, self.weight)
    346 
    347 class Conv3d(_ConvNd):

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/conv.py in conv2d_forward(self, input, weight)
    340                             _pair(0), self.dilation, self.groups)
    341         return F.conv2d(input, weight, self.bias, self.stride,
--> 342                         self.padding, self.dilation, self.groups)
    343 
    344     def forward(self, input):

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

From my understanding it seems that the model that I am importing is ready for the CPU, but I am training on the GPU, so I can I integrate the two

[muellerzr (Zachary Mueller)]

You need to change your models’ device, or your input data’s device. load_learner defaults to CPU, pass cpu=False to load your model on the GPU (what I think you’re after)

[Albertotono]

Awesome Zach, like always, let me try.
It worked

Thank you so much

[Albertotono]

It is great to share this journey here.

After this first step, I am running into my second Issue:

I have this issue

Target size (torch.Size([3, 5])) must be the same as input size (torch.Size([3, 4]))

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-26-38b1530800fe> in <module>
      1 loss_func = nn.BCEWithLogitsLoss()
----> 2 loss = loss_func(activs, y)
      3 loss

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    530             result = self._slow_forward(*input, **kwargs)
    531         else:
--> 532             result = self.forward(*input, **kwargs)
    533         for hook in self._forward_hooks.values():
    534             hook_result = hook(self, input, result)

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/modules/loss.py in forward(self, input, target)
    599                                                   self.weight,
    600                                                   pos_weight=self.pos_weight,
--> 601                                                   reduction=self.reduction)
    602 
    603 

/opt/conda/envs/fastai/lib/python3.7/site-packages/torch/nn/functional.py in binary_cross_entropy_with_logits(input, target, weight, size_average, reduce, reduction, pos_weight)
   2122 
   2123     if not (target.size() == input.size()):
-> 2124         raise ValueError("Target size ({}) must be the same as input size ({})".format(target.size(), input.size()))
   2125 
   2126     return torch.binary_cross_entropy_with_logits(input, target, weight, pos_weight, reduction_enum)

This should be caused by the fact that I have 4 category in the previous model that I trained and in this one I have 5 categories.

1. How can I check the categories? there should be a “vocab” function
2. Not sure also why they are 5, even if in the Multicategory I placed 4.

[muellerzr (Zachary Mueller)]

@Albertotono

How can I check the categories ? there should be a “vocab” function

dls.vocab I believe. You can also do dls.c to get the # of them

[Albertotono]

Awesome Zack, it was counting the " ", space in my excel. Great catch. Thank you so much.

[Albertotono]

Last step,
at the end when I am building the final learner:

learn = cnn_learner(dls, model, metrics=partial(accuracy_multi, thresh=0.2))
learn.fine_tune(3, base_lr=3e-3, freeze_epochs=4)

I used model and that model is the one that I load from my previous training:
model = load_learner(path1/‘export.pkl’, cpu=False)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-69-cb015ccf8de5> in <module>
----> 1 learn = cnn_learner(dls, model, metrics=partial(accuracy_multi, thresh=0.2))
      2 learn.fine_tune(3, base_lr=3e-3, freeze_epochs=4)

/opt/conda/envs/fastai/lib/python3.7/site-packages/fastai2/vision/learner.py in cnn_learner(dls, arch, loss_func, pretrained, cut, splitter, y_range, config, n_out, normalize, **kwargs)
    173     if normalize: _add_norm(dls, meta, pretrained)
    174     if y_range is None and 'y_range' in config: y_range = config.pop('y_range')
--> 175     model = create_cnn_model(arch, n_out, ifnone(cut, meta['cut']), pretrained, y_range=y_range, **config)
    176     learn = Learner(dls, model, loss_func=loss_func, splitter=ifnone(splitter, meta['split']), **kwargs)
    177     if pretrained: learn.freeze()

/opt/conda/envs/fastai/lib/python3.7/site-packages/fastai2/vision/learner.py in create_cnn_model(arch, n_out, cut, pretrained, n_in, init, custom_head, concat_pool, **kwargs)
     99                      concat_pool=True, **kwargs):
    100     "Create custom convnet architecture using `arch`, `n_in` and `n_out`"
--> 101     body = create_body(arch, n_in, pretrained, cut)
    102     if custom_head is None:
    103         nf = num_features_model(nn.Sequential(*body.children())) * (2 if concat_pool else 1)

/opt/conda/envs/fastai/lib/python3.7/site-packages/fastai2/vision/learner.py in create_body(arch, n_in, pretrained, cut)
     64 def create_body(arch, n_in=3, pretrained=True, cut=None):
     65     "Cut off the body of a typically pretrained `arch` as determined by `cut`"
---> 66     model = arch(pretrained=pretrained)
     67     _update_first_layer(model, n_in, pretrained)
     68     #cut = ifnone(cut, cnn_config(arch)['cut'])

TypeError: call() got an unexpected keyword argument 'pretrained’

[muellerzr (Zachary Mueller)]

@Albertotono

I used model and that model is the one that I load from my previous training:

You shouldn’t use cnn_learner anymore then. It’s not designed for this you should make a Learner instead and make sure to pass in the loss function

(You can also do splitter=default_split and then call learn.freeze() and we just mimicked cnn_learner)

[Albertotono]

@muellerzr

Learner

Thank you so much Zack, I think I am getting closer, and I am learning a lot of new things which is great. Thanks

I was looking at the documentation:
https://docs.fast.ai/basic_train.html#Learner

Learner ( data : DataBunch , model : Module , opt_func : Callable = 'Adam' , loss_func : Callable = None , metrics : Collection [ Callable ]= None , true_wd : bool = True , bn_wd : bool = True , wd : Floats = 0.01 , train_bn : bool = True , path : str = None , model_dir : PathOrStr = 'models' , callback_fns : Collection [ Callable ]= None , callbacks : Collection [ Callback ]= <factory> , layer_groups : ModuleList = None , add_time : bool = True , silent : bool = None )

model:Module?

I create this learner
learn = Learner(dls, model=model, loss_func = nn.BCEWithLogitsLoss(), metrics=partial(accuracy_multi, thresh=0.2), path=path)

but I got this message

**---------------------------------------------------------------------------**
**AttributeError                            Traceback (most recent call last)**
**<ipython-input-73-62a02725519a> in <module>**
**----> 1 learn = Learner(dls, model=model, loss_func = nn.BCEWithLogitsLoss(), metrics=partial(accuracy_multi, thresh=0.2), path=path)**

**/opt/conda/envs/fastai/lib/python3.7/site-packages/fastai2/learner.py in __init__(self, dls, model, loss_func, opt_func, lr, splitter, cbs, metrics, path, model_dir, wd, wd_bn_bias, train_bn, moms)**
**     84         self.path = Path(path) if path is not None else getattr(dls, 'path', Path('.'))**
**     85         self.add_cbs([(cb() if isinstance(cb, type) else cb) for cb in L(defaults.callbacks)+L(cbs)])**
**---> 86         self.model.to(self.dls.device)**
**     87         if hasattr(self.model, 'reset'): self.model.reset()**
**     88         self.epoch,self.n_epoch,self.loss = 0,1,tensor(0.)**

**AttributeError: 'Learner' object has no attribute 'to'**

Are there any reference or other examples to look at?

[Albertotono]

Probably I should have look at the
This documentation
https://dev.fast.ai/learner

Learner ( dls , model , loss_func = None , opt_func = 'Adam' , lr = 0.001 , 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) )

But still same issue: AttributeError: ‘Learner’ object has no attribute 'to’

[barnacl (Rekil Prashanth)]

@Albertotono

model = load_learner(path1/‘export.pkl’, cpu=False)

this is a learner object, to get your model you will ahve to do learn.model

not sure if this will work but try:

learn = load_learner(path1/'export.pkl', cpu=False)

def model_resnet_old(pretrained=True):
  return learn.model

learn = cnn_learner(dls, model_resnet_old, metrics=partial(accuracy_multi, thresh=0.2))
learn.fine_tune(3, base_lr=3e-3, freeze_epochs=4)

again i haven’t tried this
if you could share a link to the notebook will be easier to test

[muellerzr (Zachary Mueller)]

D’oh! @Albertotono @barnacl is definitely on the right track. You can try the cnn_learner way but if that doesn’t work, on the call to Learner use your learn.model from your loaded Learner (see his resnet_old code)

[Albertotono]

@barnacl

learn.model

Dear @barnacl it worked

Thank you so much guys, @muellerzr @barnacl

[Albertotono]

Just as FYI

Performance with Resnet50

Performance with my previous model trained on the same objects (categories) but with ResNet34

This is the model that I have imported

Probably I will need to perform an Object Detection to better perform this task.
And also I should probably use Pretrained=False .