How to set file path

bvb · August 27, 2020, 12:57am

learn = cnn_learner(data, resnet34, metrics=error_rate)

I want to load the model with the file I have dowloaded from the website,
but I don’t konw whether the cnn_learner function has a parameter to set the file path

ali_baba · August 27, 2020, 1:28am

you can pass a path to the learner as such:

learn = cnn_learner(data, resnet34, path=path, metrics=error_rate)

To see the possible parameters in a learner run the following:
??cnn_learner

Now what this path is pointing to is a different story. I am using my local machine and therefore I like setting the file path to somewhere I can easily navigate to with the GUI – mainly to delete things that I do terribly

immarried · August 27, 2020, 4:29am

Check if there’s a load method, learn.load, that you can use.

github.com

qAp/fastai/blob/master/fastai/learner.py#L287


        if hasattr(self.loss_func, 'reduction'): return replacing_yield(self.loss_func, 'reduction', 'none')
        else: return replacing_yield(self, 'loss_func', partial(self.loss_func, reduction='none'))

    @delegates(save_model)
    def save(self, file, **kwargs):
        file = join_path_file(file, self.path/self.model_dir, ext='.pth')
        save_model(file, self.model, getattr(self,'opt',None), **kwargs)
        return file

    @delegates(load_model)
    def load(self, file, with_opt=None, device=None, **kwargs):
        if device is None and hasattr(self.dls, 'device'): device = self.dls.device
        if self.opt is None: self.create_opt()
        file = join_path_file(file, self.path/self.model_dir, ext='.pth')
        load_model(file, self.model, self.opt, device=device, **kwargs)
        return self

Learner.x,Learner.y = add_props(lambda i,x: detuplify((x.xb,x.yb)[i]))

# Cell
add_docs(Learner, "Group together a `model`, some `dls` and a `loss_func` to handle training",

You need to set learn.path and learn.model_dir such that they allow you to point to the file containing the model you downloaded.

bvb · August 27, 2020, 3:27pm

I hava downloaded the file resnet34-333f7ec4.pth to the path = ‘/Users/lc/Desktop/test/’

learn = cnn_learner(dls, resnet34, metrics = error_rate, path = path)

But it still downloads the file resnet34-333f7ec4.pth.

ali_baba · August 27, 2020, 11:41pm

Are you using a local machine or working off a server?

I am a bit confused as to what you’re trying to do, are you trying to upload (unique) trained weights into a resnet34 architecture? the way you are calling it there – the cnn_learner call will be instantiated with the trained weights for imagenet.

@immarried’s suggestion should work. You could instantiate it with normal imagenet weights like you are trying and then call learn.load(x) where x would be the file containing the weights you want.

This also seems like a promising route to try: Loading pretrained weights that are not from ImageNet