# This Python 3 environment comes with many helpful analytics libraries installed
# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python
# For example, here's several helpful packages to load

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)

# Input data files are available in the read-only "../input/" directory
# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory

import os
for dirname, _, filenames in os.walk('/kaggle/input'):
    for filename in filenames:
        print(os.path.join(dirname, filename))

# You can write up to 20GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using "Save & Run All" 
# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session

# Use the kagglehub client library to attach Kaggle resources like competitions, datasets, and models to your session
# Learn more about kagglehub: https://github.com/Kaggle/kagglehub/blob/main/README.md

import kagglehub
# kagglehub.dataset_download('<owner>/<dataset-slug>')


get_ipython().getoutput("pip install git+https://github.com/jacobgil/pytorch-grad-cam.git -q")


import matplotlib.pyplot as plt
import seaborn as sns
from PIL import Image
import copy
import os
import shutil
import random

from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, classification_report, confusion_matrix , f1_score

import torch
import torch.nn as nn
import torch.optim as optim
import torch.cuda.amp as amp  
import torchvision.models as models
import torchvision.transforms as transforms
import torchvision.datasets as datasets
from torchvision.datasets import ImageFolder
from torch.utils.data import DataLoader ,Dataset
from torch.optim.lr_scheduler import ReduceLROnPlateau
from tqdm import tqdm
from tqdm.auto import tqdm
from torchvision.models import efficientnet_b3  
from pytorch_grad_cam import GradCAM
from pytorch_grad_cam.utils.image import show_cam_on_image

import warnings
warnings.filterwarnings("ignore")


original_dirs = {
    'Calculus': '/kaggle/input/datasets/salmansajid05/oral-diseases/Calculus/Calculus',
    'Caries': '/kaggle/input/datasets/salmansajid05/oral-diseases/Data caries/Data caries/caries augmented data set/preview',
    'Gingivitis': '/kaggle/input/datasets/salmansajid05/oral-diseases/Gingivitis/Gingivitis',
    'Ulcers': '/kaggle/input/datasets/salmansajid05/oral-diseases/Mouth Ulcer/Mouth Ulcer/Mouth_Ulcer_augmented_DataSet/preview',
    'Tooth Discoloration': '/kaggle/input/datasets/salmansajid05/oral-diseases/Tooth Discoloration/Tooth Discoloration /Tooth_discoloration_augmented_dataser/preview',
    'Hypodontia': '/kaggle/input/datasets/salmansajid05/oral-diseases/hypodontia/hypodontia'
}


DATA_DIR = "/kaggle/working/oral_dataset"
splits = ['train', 'val', 'test']
classes = list(original_dirs.keys())
classes


import os
import shutil

NEW_DATASET = "/kaggle/working/oral_dataset"

os.makedirs(NEW_DATASET, exist_ok=True)

for class_name, source_dir in original_dirs.items():

    target_dir = os.path.join(NEW_DATASET, class_name)
    os.makedirs(target_dir, exist_ok=True)

    for file in os.listdir(source_dir):
        if file.lower().endswith((".jpg", ".jpeg", ".png")):
            shutil.copy(
                os.path.join(source_dir, file),
                os.path.join(target_dir, file)
            )

print("Done")


full_dataset = datasets.ImageFolder(root=DATA_DIR)

CLASS_NAMES = full_dataset.classes
NUM_CLASSES = len(CLASS_NAMES)


DATA_DIR = original_dirs  

CHECKPOINT_DIR = "./checkpoints"     
OUTPUT_DIR = "./outputs"           
os.makedirs(CHECKPOINT_DIR, exist_ok=True)
os.makedirs(OUTPUT_DIR, exist_ok=True)

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")

IMG_SIZE = 224
BATCH_SIZE = 32
NUM_WORKERS = 4
TRAIN_RATIO, VAL_RATIO, TEST_RATIO = 0.8, 0.1, 0.1
SEED = 42

EPOCHS = 30                      
LR_SCRATCH = 1e-3                 
LR_PRETRAINED_HEAD = 1e-3          
LR_PRETRAINED_FINETUNE = 1e-5     

WEIGHT_DECAY = 1e-4        
LABEL_SMOOTHING = 0.1      
DROPOUT = 0.4

EARLY_STOPPING_PATIENCE = 2
FREEZE_EPOCHS = 5              

torch.manual_seed(SEED)
print("Done")


IMAGENET_MEAN = [0.485, 0.456, 0.406]
IMAGENET_STD = [0.229, 0.224, 0.225]

train_transform = transforms.Compose([
    transforms.Resize((IMG_SIZE + 20, IMG_SIZE + 20)),
    transforms.RandomResizedCrop(IMG_SIZE, scale=(0.8, 1.0)),
    transforms.RandomHorizontalFlip(p=0.5),
    transforms.RandomRotation(degrees=15),
    transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2),
    transforms.ToTensor(),
    transforms.Normalize(IMAGENET_MEAN, IMAGENET_STD),
    transforms.RandomErasing(p=0.2),
])

eval_transform = transforms.Compose([
    transforms.Resize((IMG_SIZE, IMG_SIZE)),
    transforms.ToTensor(),
    transforms.CenterCrop(224),
    transforms.Normalize(IMAGENET_MEAN, IMAGENET_STD),
])



class SubsetWithTransform(Dataset):
    def __init__(self, base_dataset, indices, transform):
        self.base_dataset = base_dataset
        self.indices = indices
        self.transform = transform

    def __len__(self):
        return len(self.indices)

    def __getitem__(self, idx):
        real_idx = self.indices[idx]
        path, label = self.base_dataset.samples[real_idx]
        image = self.base_dataset.loader(path)
        image = self.transform(image)
        return image, label


targets = np.array([label for _, label in full_dataset.samples])
indices = np.arange(len(full_dataset))

train_idx, temp_idx = train_test_split(
    indices, test_size=(1 - TRAIN_RATIO), stratify=targets, random_state=SEED
)

val_ratio_of_temp = VAL_RATIO / (VAL_RATIO + TEST_RATIO)
val_idx, test_idx = train_test_split(
    temp_idx, test_size=(1 - val_ratio_of_temp),
    stratify=targets[temp_idx], random_state=SEED
)

train_ds = SubsetWithTransform(full_dataset, train_idx, train_transform)
val_ds = SubsetWithTransform(full_dataset, val_idx, eval_transform)
test_ds = SubsetWithTransform(full_dataset, test_idx, eval_transform)

train_loader = DataLoader(
    train_ds,
    batch_size=BATCH_SIZE,
    shuffle=True,
    num_workers=0,
    pin_memory=True,
    drop_last=True
)

val_loader = DataLoader(
    val_ds,
    batch_size=BATCH_SIZE,
    shuffle=False,
    num_workers=0,
    pin_memory=True
)

test_loader = DataLoader(
    test_ds,
    batch_size=BATCH_SIZE,
    shuffle=False,
    num_workers=0,
    pin_memory=True
)

print(f"Count classes: {NUM_CLASSES} -> {CLASS_NAMES}")
print(f"Count image of train: {len(train_ds)}")
print(f"Count image of Validation: {len(val_ds)}")
print(f"Test: {len(test_ds)}")


fig, axes = plt.subplots(2, 3, figsize=(10, 7))
for ax in axes.flatten():
    img, label = train_ds[np.random.randint(len(train_ds))]
    img = img * torch.tensor(IMAGENET_STD).view(3,1,1) + torch.tensor(IMAGENET_MEAN).view(3,1,1)
    img = img.clamp(0, 1).permute(1, 2, 0).numpy()
    ax.imshow(img)
    ax.set_title(CLASS_NAMES[label], fontsize=9)
    ax.axis("off")
plt.tight_layout()
plt.show()


class ResidualBlock(nn.Module):
    def __init__(self, in_channels, out_channels, stride=1):
        super().__init__()
        self.conv1 = nn.Conv2d(in_channels, out_channels, 3, stride, 1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channels)
        self.conv2 = nn.Conv2d(out_channels, out_channels, 3, 1, 1, bias=False)
        self.bn2 = nn.BatchNorm2d(out_channels)
        self.relu = nn.ReLU(inplace=True)

        self.shortcut = nn.Sequential()
        if stride != 1 or in_channels != out_channels:
            self.shortcut = nn.Sequential(
                nn.Conv2d(in_channels, out_channels, 1, stride, bias=False),
                nn.BatchNorm2d(out_channels),
            )

    def forward(self, x):
        identity = self.shortcut(x)
        out = self.relu(self.bn1(self.conv1(x)))
        out = self.bn2(self.conv2(out))
        out += identity          # skip connection
        return self.relu(out)


class ScratchCNN(nn.Module):
    def __init__(self, num_classes, dropout=DROPOUT):
        super().__init__()
        self.stem = nn.Sequential(
            nn.Conv2d(3, 64, 7, 2, 3, bias=False),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(3, 2, 1),
        )
        self.stage1 = self._make_stage(64, 64, 2, stride=1)
        self.stage2 = self._make_stage(64, 128, 2, stride=2)
        self.stage3 = self._make_stage(128, 256, 2, stride=2)
        self.stage4 = self._make_stage(256, 512, 2, stride=2)

        self.global_pool = nn.AdaptiveAvgPool2d(1)
        self.dropout = nn.Dropout(dropout)
        self.classifier = nn.Linear(512, num_classes)

    def _make_stage(self, in_c, out_c, num_blocks, stride):
        layers = [ResidualBlock(in_c, out_c, stride)]
        for _ in range(num_blocks - 1):
            layers.append(ResidualBlock(out_c, out_c, 1))
        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.stem(x)
        x = self.stage1(x); x = self.stage2(x)
        x = self.stage3(x); x = self.stage4(x)
        x = self.global_pool(x)
        x = torch.flatten(x, 1)
        x = self.dropout(x)
        return self.classifier(x)

_test_out = ScratchCNN(num_classes=NUM_CLASSES)(torch.randn(2, 3, IMG_SIZE, IMG_SIZE))


def build_resnet50(num_classes, dropout=DROPOUT):
    m = models.resnet50(weights=models.ResNet50_Weights.IMAGENET1K_V2)
    m.fc = nn.Sequential(nn.Dropout(dropout), nn.Linear(m.fc.in_features, num_classes))
    return m

def build_efficientnet_b0(num_classes, dropout=DROPOUT):
    m = models.efficientnet_b0(weights=models.EfficientNet_B0_Weights.IMAGENET1K_V1)
    m.classifier = nn.Sequential(nn.Dropout(dropout), nn.Linear(m.classifier[1].in_features, num_classes))
    return m

def build_densenet121(num_classes, dropout=DROPOUT):
    m = models.densenet121(weights=models.DenseNet121_Weights.IMAGENET1K_V1)
    m.classifier = nn.Sequential(nn.Dropout(dropout), nn.Linear(m.classifier.in_features, num_classes))
    return m


def freeze_backbone(model, model_name):
    for p in model.parameters():
        p.requires_grad = False
    head = model.fc if model_name == "resnet50" else model.classifier
    for p in head.parameters():
        p.requires_grad = True
    return model

def unfreeze_all(model):
    for p in model.parameters():
        p.requires_grad = True
    return model

def count_parameters(model):
    return sum(p.numel() for p in model.parameters() if p.requires_grad)


class EarlyStopping:
    def __init__(self, patience=EARLY_STOPPING_PATIENCE):
        self.patience = patience
        self.best_loss = float("inf")
        self.counter = 0
        self.best_state = None
        self.should_stop = False

    def step(self, val_loss, model):
        if val_loss < self.best_loss:
            self.best_loss = val_loss
            self.counter = 0
            self.best_state = copy.deepcopy(model.state_dict())
        else:
            self.counter += 1
            if self.counter >= self.patience:
                self.should_stop = True


def run_one_epoch(model, loader, criterion, optimizer=None):
    is_training = optimizer is not None
    model.train() if is_training else model.eval()

    total_loss, total_correct, total_samples = 0.0, 0, 0
    torch.set_grad_enabled(is_training)
    for images, labels in tqdm(loader, leave=False):
        images, labels = images.to(DEVICE), labels.to(DEVICE)
        if is_training:
            optimizer.zero_grad()
        outputs = model(images)
        loss = criterion(outputs, labels)
        if is_training:
            loss.backward()
            optimizer.step()
        total_loss += loss.item() * images.size(0)
        total_correct += (outputs.argmax(1) == labels).sum().item()
        total_samples += images.size(0)
    torch.set_grad_enabled(True)

    return total_loss / total_samples, total_correct / total_samples


def train_model(model, model_name, train_loader, val_loader, epochs, lr,
                 weight_decay=WEIGHT_DECAY, label_smoothing=LABEL_SMOOTHING):
    model = model.to(DEVICE)
    criterion = nn.CrossEntropyLoss(label_smoothing=label_smoothing)
    optimizer = torch.optim.AdamW(
        filter(lambda p: p.requires_grad, model.parameters()),
        lr=lr, weight_decay=weight_decay,
    )
    scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode="min", factor=0.5, patience=2)
    early_stopper = EarlyStopping()

    history = {"train_loss": [], "train_acc": [], "val_loss": [], "val_acc": []}

    for epoch in range(1, epochs + 1):
        train_loss, train_acc = run_one_epoch(model, train_loader, criterion, optimizer)
        val_loss, val_acc = run_one_epoch(model, val_loader, criterion, optimizer=None)
        scheduler.step(val_loss)

        history["train_loss"].append(train_loss)
        history["train_acc"].append(train_acc)
        history["val_loss"].append(val_loss)
        history["val_acc"].append(val_acc)

        print(f"[{model_name}] Epoch {epoch}/{epochs} | "
              f"train_loss={train_loss:.4f} train_acc={train_acc:.4f} | "
              f"val_loss={val_loss:.4f} val_acc={val_acc:.4f}")

        early_stopper.step(val_loss, model)
        if early_stopper.should_stop:
            print(f"[{model_name}] Early stopping {epoch}")
            break

    model.load_state_dict(early_stopper.best_state)  
    return model, history


@torch.no_grad()
def evaluate_full(model, loader):
    model.eval()
    all_preds, all_labels = [], []
    for images, labels in tqdm(loader, leave=False, desc="Evaluating"):
        images = images.to(DEVICE)
        preds = model(images).argmax(1).cpu().numpy()
        all_preds.extend(preds)
        all_labels.extend(labels.numpy())

    acc = accuracy_score(all_labels, all_preds)
    f1 = f1_score(all_labels, all_preds, average="macro")
    return {"accuracy": acc, "f1_macro": f1, "predictions": all_preds, "labels": all_labels}

print("Done")


def plot_history(history, model_name):
    fig, axes = plt.subplots(1, 2, figsize=(12, 4))
    axes[0].plot(history["train_loss"], label="Train Loss")
    axes[0].plot(history["val_loss"], label="Val Loss")
    axes[0].set_title(f"{model_name} - Loss"); axes[0].set_xlabel("Epoch"); axes[0].legend()

    axes[1].plot(history["train_acc"], label="Train Acc")
    axes[1].plot(history["val_acc"], label="Val Acc")
    axes[1].set_title(f"{model_name} - Accuracy"); axes[1].set_xlabel("Epoch"); axes[1].legend()

    plt.tight_layout()
    plt.savefig(os.path.join(OUTPUT_DIR, f"{model_name}_history.png"), dpi=150)
    plt.show()


def plot_confusion_matrix(labels, preds, class_names, model_name):
    cm = confusion_matrix(labels, preds)
    fig, ax = plt.subplots(figsize=(7, 6))
    im = ax.imshow(cm, cmap="Blues")
    ax.set_xticks(range(len(class_names))); ax.set_yticks(range(len(class_names)))
    ax.set_xticklabels(class_names, rotation=45, ha="right")
    ax.set_yticklabels(class_names)
    ax.set_xlabel("Predicted"); ax.set_ylabel("True")
    ax.set_title(f"Confusion Matrix - {model_name}")
    for i in range(len(class_names)):
        for j in range(len(class_names)):
            ax.text(j, i, cm[i, j], ha="center", va="center",
                     color="white" if cm[i, j] > cm.max()/2 else "black")
    plt.colorbar(im)
    plt.tight_layout()
    plt.savefig(os.path.join(OUTPUT_DIR, f"{model_name}_confusion_matrix.png"), dpi=150)
    plt.show()


scratch_model = ScratchCNN(NUM_CLASSES)
print(f"{count_parameters(scratch_model):,}")

scratch_model, scratch_history = train_model(
    scratch_model, "scratch_cnn", train_loader, val_loader,
    epochs=EPOCHS, lr=LR_SCRATCH,
)


plot_history(scratch_history, "scratch_cnn")


resnet_model = build_resnet50(NUM_CLASSES)
resnet_model = freeze_backbone(resnet_model, "resnet50")

resnet_model, resnet_history_1 = train_model(
    resnet_model, "resnet50", train_loader, val_loader,
    epochs=FREEZE_EPOCHS, lr=LR_PRETRAINED_HEAD,
)


resnet_model = unfreeze_all(resnet_model)
resnet_model, resnet_history_2 = train_model(
    resnet_model, "resnet50", train_loader, val_loader,
    epochs=max(EPOCHS - FREEZE_EPOCHS, 5), lr=LR_PRETRAINED_FINETUNE,
)

resnet_history = {k: resnet_history_1[k] + resnet_history_2[k] for k in resnet_history_1}
plot_history(resnet_history, "resnet50")


effnet_model = build_efficientnet_b0(NUM_CLASSES)
effnet_model = freeze_backbone(effnet_model, "efficientnet_b0")

effnet_model, effnet_history_1 = train_model(
    effnet_model, "efficientnet_b0", train_loader, val_loader,
    epochs=FREEZE_EPOCHS, lr=LR_PRETRAINED_HEAD,
)


effnet_model = unfreeze_all(effnet_model)
effnet_model, effnet_history_2 = train_model(
    effnet_model, "efficientnet_b0", train_loader, val_loader,
    epochs=max(EPOCHS - FREEZE_EPOCHS, 5), lr=LR_PRETRAINED_FINETUNE,
)

effnet_history = {k: effnet_history_1[k] + effnet_history_2[k] for k in effnet_history_1}
plot_history(effnet_history, "efficientnet_b0")


densenet_model = build_densenet121(NUM_CLASSES)
densenet_model = freeze_backbone(densenet_model, "densenet121")

densenet_model, densenet_history_1 = train_model(
    densenet_model, "densenet121", train_loader, val_loader,
    epochs=FREEZE_EPOCHS, lr=LR_PRETRAINED_HEAD,
)


densenet_model = unfreeze_all(densenet_model)
densenet_model, densenet_history_2 = train_model(
    densenet_model, "densenet121", train_loader, val_loader,
    epochs=max(EPOCHS - FREEZE_EPOCHS, 5), lr=LR_PRETRAINED_FINETUNE,
)

densenet_history = {k: densenet_history_1[k] + densenet_history_2[k] for k in densenet_history_1}
plot_history(densenet_history, "densenet121")


all_models = {
    "scratch_cnn": scratch_model,
    "resnet50": resnet_model,
    "efficientnet_b0": effnet_model,
    "densenet121": densenet_model,
}

results = []
best_model_name, best_f1 = None, -1.0

for name, model in all_models.items():
    print(f"Evaluate {name} on test set...")
    eval_result = evaluate_full(model, test_loader)

    ckpt_path = os.path.join(CHECKPOINT_DIR, f"{name}.pth")
    torch.save(model.state_dict(), ckpt_path)
    print(f"The save weight {name} in: {ckpt_path}")

    plot_confusion_matrix(eval_result["labels"], eval_result["predictions"], CLASS_NAMES, name)

    results.append({
        "model": name,
        "trainable_params": count_parameters(model),
        "test_accuracy": round(eval_result["accuracy"], 4),
        "test_f1": round(eval_result["f1_macro"], 4),
    })

    if eval_result["f1_macro"] > best_f1:
        best_f1 = eval_result["f1_macro"]
        best_model_name = name

comparison_df = pd.DataFrame(results).sort_values("test_f1", ascending=False).reset_index(drop=True)
comparison_df.to_csv(os.path.join(OUTPUT_DIR, "models_comparison.csv"), index=False)
comparison_df


for name, model in all_models.items():
    print(f"Evaluate {name} on test set...")

    eval_result = evaluate_full(model, test_loader)

    print(f"Classification Report for {name}")
    print(
        classification_report(
            eval_result["labels"],
            eval_result["predictions"],
            target_names=CLASS_NAMES,
            digits=4
        )
    )

    ckpt_path = os.path.join(CHECKPOINT_DIR, f"{name}.pth")
    torch.save(model.state_dict(), ckpt_path)


best_model = all_models[best_model_name]
best_path = os.path.join(CHECKPOINT_DIR, "best_model.pth")

torch.save({
    "model_name": best_model_name,
    "state_dict": best_model.state_dict(),
    "class_names": CLASS_NAMES,
    "test_f1": best_f1,
}, best_path)

print(f"The Best Model: {best_model_name}  (F1 = {best_f1:.4f})")
print(f"Save: {best_path}")


IMAGE_PATH = '/kaggle/working/oral_dataset/Hypodontia/(100).JPG'

checkpoint = torch.load(os.path.join(CHECKPOINT_DIR, "best_model.pth"), map_location=DEVICE)
loaded_name = checkpoint["model_name"]
loaded_classes = checkpoint["class_names"]

builders = {
    "scratch_cnn": lambda: ScratchCNN(len(loaded_classes)),
    "resnet50": lambda: build_resnet50(len(loaded_classes)),
    "efficientnet_b0": lambda: build_efficientnet_b0(len(loaded_classes)),
    "densenet121": lambda: build_densenet121(len(loaded_classes)),
}
inference_model = builders[loaded_name]()
inference_model.load_state_dict(checkpoint["state_dict"])
inference_model.to(DEVICE).eval()
print(f"Download the best model: {loaded_name} (test F1 = {checkpoint['test_f1']:.4f})")

with torch.no_grad():
    image = Image.open(IMAGE_PATH).convert("RGB")
    tensor = eval_transform(image).unsqueeze(0).to(DEVICE)
    probs = torch.softmax(inference_model(tensor), dim=1)[0]
    pred_idx = probs.argmax().item()

plt.imshow(image); plt.axis("off")
plt.title(f"Classifier: {loaded_classes[pred_idx]} ({probs[pred_idx]*100:.1f}%)")
plt.show()

for cname, p in sorted(zip(loaded_classes, probs.tolist()), key=lambda x: -x[1]):
    print(f"  {cname:25s}: {p*100:5.2f}%")


IMAGE_PATH = '/kaggle/working/oral_dataset/Caries/caries_0_1001.jpeg'

checkpoint = torch.load(os.path.join(CHECKPOINT_DIR, "best_model.pth"), map_location=DEVICE)
loaded_name = checkpoint["model_name"]
loaded_classes = checkpoint["class_names"]

builders = {
    "scratch_cnn": lambda: ScratchCNN(len(loaded_classes)),
    "resnet50": lambda: build_resnet50(len(loaded_classes)),
    "efficientnet_b0": lambda: build_efficientnet_b0(len(loaded_classes)),
    "densenet121": lambda: build_densenet121(len(loaded_classes)),
}
inference_model = builders[loaded_name]()
inference_model.load_state_dict(checkpoint["state_dict"])
inference_model.to(DEVICE).eval()
print(f"Download the best model: {loaded_name} (test F1 = {checkpoint['test_f1']:.4f})")

with torch.no_grad():
    image = Image.open(IMAGE_PATH).convert("RGB")
    tensor = eval_transform(image).unsqueeze(0).to(DEVICE)
    probs = torch.softmax(inference_model(tensor), dim=1)[0]
    pred_idx = probs.argmax().item()

plt.imshow(image); plt.axis("off")
plt.title(f"Classifier: {loaded_classes[pred_idx]} ({probs[pred_idx]*100:.1f}%)")
plt.show()

for cname, p in sorted(zip(loaded_classes, probs.tolist()), key=lambda x: -x[1]):
    print(f"  {cname:25s}: {p*100:5.2f}%")


import shutil

shutil.make_archive(
    "/kaggle/working/output_files",
    "zip",
    "/kaggle/working/"
)


from IPython.display import FileLink

FileLink('/kaggle/working/output_files.zip')