sadimanna

class GradCamModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.gradients = None
        self.tensorhook = []
        self.layerhook = []
        self.selected_out = None
        
        #PRETRAINED MODEL
        self.pretrained = models.resnet50(pretrained=True)

sadimanna

fig,ax = plt.subplots(1,1, figsize=(10,10))

b = []
for i in range(num_plyrs):
    b.append(ax.bar(x - (i - num_plyrs/2 + 0.5)*w, 
                    stats.loc[i].values[1:], 
                    width=w, 
                    color=colors(i), 
                    align='center', 
                    edgecolor = 'black', 

sadimanna

tdg = DSDataGen('test', testimages, testlabels, num_classes=10)

tdl = DataLoader(tdg, batch_size = 32, drop_last = True)

dsmodel.eval()
    
loss_sublist = np.array([])
acc_sublist = np.array([])

with torch.no_grad():

sadimanna

for epoch in range(20):
    
    stime = time.time()
    print("=============== Epoch : %3d ==============="%(epoch+1))
    
    loss_sublist = np.array([])
    acc_sublist = np.array([])
    
    #iter_num = 0
    dsmodel.train()

sadimanna

tr_ep_loss = []
tr_ep_acc = []

val_ep_loss = []
val_ep_acc = []

min_val_loss = 100.0

EPOCHS = 10
num_cl = 10

sadimanna

class DSDataGen(Dataset):
    def __init__(self, phase, imgarr,labels,num_classes):
        
        self.phase = phase
        self.num_classes = num_classes
        self.imgarr = imgarr
        self.labels = labels
        
        self.randomcrop = transforms.RandomResizedCrop(32,(0.8,1.0))
    

sadimanna

class DSModel(nn.Module):
    def __init__(self,premodel,num_classes):
        super().__init__()
        
        self.premodel = premodel
        self.num_classes = num_classes
        
        for p in self.premodel.parameters():
            p.requires_grad = False
            

sadimanna

nr = 0
current_epoch = 0
epochs = 100
tr_loss = []
val_loss = []

for epoch in range(100):
        
    print(f"Epoch [{epoch}/{epochs}]\t")
    stime = time.time()

sadimanna

def save_model(model, optimizer, scheduler, current_epoch, name):
    out = os.path.join('/content/saved_models/',name.format(current_epoch))

    torch.save({'model_state_dict': model.state_dict(),
                'optimizer_state_dict': optimizer.state_dict(),
                'scheduler_state_dict':scheduler.state_dict()}, out)

def plot_features(model, num_classes, num_feats, batch_size):
    preds = np.array([]).reshape((0,1))
    gt = np.array([]).reshape((0,1))

sadimanna

#OPTMIZER
optimizer = LARS(
    [params for params in model.parameters() if params.requires_grad],
    lr=0.2,
    weight_decay=1e-6,
    exclude_from_weight_decay=["batch_normalization", "bias"],
)

# "decay the learning rate with the cosine decay schedule without restarts"
#SCHEDULER OR LINEAR EWARMUP