Return value from custom function in Dataset class

data
1 
class Car_insurance(Dataset):
    def __init__(self, encoding = "label_encode", embedding_layer = False, normal = False):
        path = '../../data/insurance_claims.csv'
        categorical_cols = ['policy_state', 'umbrella_limit', 'insured_sex', 'insured_education_level',
    	'insured_occupation', 'insured_hobbies', 'insured_relationship', 'incident_type',
        'collision_type', 'incident_severity', 'authorities_contacted', 'incident_state', 'incident_city',	
        'property_damage', 'police_report_available', 'auto_make', 'auto_model']
        cols_to_remove = ['policy_number', 'policy_bind_date', 'policy_csl', 'incident_location', 'incident_date', '_c39']
        
        self.normal = normal
        data = load_data(path)
        data = remove_cols(data, cols_to_remove)
        if normal:
            data = get_normal_data(data, "car_insurance")
        self.label = get_labels(data, "car_insurance")
        self.cat_cols = data[categorical_cols]
        self.cont_cols = data.drop(categorical_cols, axis=1)
        if embedding_layer:
            self.cat_cols = self.cat_cols.astype("category")
            self.embedded_cols = {n: len(col.cat.categories) for n,col in self.cat_cols.items() if (col.dtype == "category")}
            self.embedding_sizes = [(n_categories, min(50, (n_categories+1)//2)) for _,n_categories in self.embedded_cols.items()]
            embedded_col_names = self.embedded_cols.keys()
            embed = []
            for i, name in enumerate(embedded_col_names):
                embed_elem = {cat : n for n, cat in enumerate(self.cat_cols[name].cat.categories)}
                embed.append(embed_elem)
                self.cat_cols[name] = self.cat_cols[name].replace(embed_elem)
            
        else:
            if encoding == 'one_hot':
                self.cat_cols = one_hot_encoding(self.cat_cols, categorical_cols)
            if encoding == 'label_encode':
                self.cat_cols = label_encoding(self.cat_cols, categorical_cols)
            if encoding == 'gel_encode':
                self.cat_cols = gel_encoding(self.cat_cols, categorical_cols)
    
    def __len__(self):
        return(len(self.label))
    
    def __getitem__(self, idx):    
        cat_cols = (self.cat_cols.values.astype(np.float32))
        cont_cols = (self.cont_cols.values.astype(np.float32))
        label = (self.label.astype(np.int32))
        return (cont_cols[idx], cat_cols[idx], label[idx])

This is my dataset class. I want to add a custom function to return self.embedding_sizes. I want self.embedding sizes before I get the items in dataloader. Please help me with the issue. I am not sure if it is possible or should I try some other way?

2

I’m not sure I understand the issue correctly.
self.embedding_sizes is currently initialized in __init__ if embedding_layer is set.
If you want to get its value, you should be able to directly access it (once set):

dataset = Car_insurance(...)
dataset.embedding_sizes

Alternatively you could also write a custom “getter” function, if that’s more convenient.

3

Thank You @ptrblck, dataset.embedding_sizes return the values exactly as I wanted. I did not know that I could access class variables with the object.

I had tried writing a custom “getter” function earlier but did not get the output. So, I thought it might not be possible. But dataset.embedding_sizes is much better. :smiley: