Cfrgtkky

This part of class i did not understand what does do in this code:

for file in os.listdir(path):

    if(os.path.isfile(os.path.join(path,file)) and select in file): 

        temp = scipy.io.loadmat(os.path.join(path,file))

        temp = {k:v for k, v in temp.items() if k[0] != '_'}

        for i  in range(len(temp[patch_type+"_patches"])):

            self.tensors.append(temp[patch_type+"_patches"][i])

            self.labels.append(temp[patch_type+"_labels"][0][i])



self.tensors = np.array(self.tensors)

self.labels = np.array(self.labels)

especially this line :

temp = {k:v for k, v in temp.items() if k[0] != '_'}

the whole class is as follow :

class Datasets(Dataset):

    def __init__(self,path,train,transform=None):

        if(train):

            select ="Training"

            patch_type = "train"

        else:

            select = "Testing"

            patch_type = "testing"



        self.tensors = 

        self.labels = 

        self.transform = transform





        for file in os.listdir(path):

            if(os.path.isfile(os.path.join(path,file)) and select in file): 



                temp = scipy.io.loadmat(os.path.join(path,file))

                temp = {k:v for k, v in temp.items() if k[0] != '_'}

                for i  in range(len(temp[patch_type+"_patches"])):

                    self.tensors.append(temp[patch_type+"_patches"][i])

                    self.labels.append(temp[patch_type+"_labels"][0][i])



        self.tensors = np.array(self.tensors)

        self.labels = np.array(self.labels)



    def __len__(self):

        try:

            if len(self.tensors) != len(self.labels):

                raise Exception("Lengths of the tensor and labels list are not the same")

        except Exception as e:

            print(e.args[0])

        return len(self.tensors)



    def __getitem__(self,idx):

        sample = (self.tensors[idx],self.labels[idx])

       # print(self.labels)

        sample = (torch.from_numpy(self.tensors[idx]),torch.from_numpy(np.array(self.labels[idx])).long())

        return sample

    #tuple containing the image patch and its corresponding label

It's a dict comprehension; in this particular case, it creates a new dict from an existing dict temp, but only for items for which the key k does not start with an underscore. That check is performed by the if ... part.

It is equivalent to

new = {}

for k, v in temp.items():

    if key[0] != '_':

        new[k] = value

temp = new

or, slightly different:

new = {}

for key, value in temp.items():

    if not key.startswith('_'):

        new[key] = value

temp = new

You can see that it looks a bit nicer as a single line, since it avoids a temporary dict (new; under the hood, it still creates a nameless temporary dict though).

It is filtering out the underscore-prefixed variables from the loaded MATLAB file. From the scipy documentation the function scipy.io.loadmat returns a dictionary containing the variable names from the loaded file as keys and the matricies as values. The line of code you reference is a dictionary comprehension that clones the dictionary minus the variables that fail the conditional check.

Update

What happens here is roughly this:

1. Load a MATLAB file (file in your code) as a hashmap (dictionary) where the keys are the variable names from the file and the values are the matricies, assign to temp.


2. Iterate through those key/value pairs and drop the underscore-prefixed ones and reassign the results of that iteration to temp.


3. Profit