import numpy as np
import torch
from pathlib import Path
import torch.nn as nn
import torch.nn.functional as F
from PIL import Image
from torchvision import transforms
import gradio as gr


transform = transforms.Compose([
    transforms.Resize((28, 28)),
    transforms.Grayscale(),
    transforms.ToTensor()
])
labels = ["๐ (ศูนย์)", "๑ (หนึ่ง)", "๒ (สอง)", "๓ (สาม)", "๔ (สี่)", "๕ (ห้า)", "๖ (หก)", "๗ (เจ็ด)", "๘ (แปด)", "๙ (เก้า)"]
LABELS = {i:k for i, k in enumerate(labels)} # dictionary of index and label


# Load model using DropoutThaiDigit instead
class DropoutThaiDigit(nn.Module):
    def __init__(self):
        super(DropoutThaiDigit, self).__init__()
        self.fc1 = nn.Linear(28 * 28, 392)
        self.fc2 = nn.Linear(392, 196)
        self.fc3 = nn.Linear(196, 98)
        self.fc4 = nn.Linear(98, 10)
        self.dropout = nn.Dropout(0.1)
        
    def forward(self, x):
        x = x.view(-1, 28 * 28)
        x = self.fc1(x)
        x = F.relu(x)
        x = self.dropout(x)
        x = self.fc2(x)
        x = F.relu(x)
        x = self.dropout(x)
        x = self.fc3(x)
        x = F.relu(x)
        x = self.dropout(x)
        x = self.fc4(x)
        return x


model = DropoutThaiDigit()
model.load_state_dict(torch.load("thai_digit_net.pth"))
model.eval()


def predict(img):
    """
    Predict function takes image and return top 5 predictions
    as a dictionary:

        {label: confidence, label: confidence, ...}
    """
    if img is None:
        return None
    img = transform(img)  # do not need to use 1 - transform(img) because gradio already do it
    probs = model(img).softmax(dim=1).ravel()
    probs, indices = torch.topk(probs, 5)  # select top 5
    probs, indices = probs.tolist(), indices.tolist()  # transform to list
    confidences = {LABELS[i]: v for i, v in zip(indices, probs)}
    return confidences


gr.Interface(
    fn=predict,
    inputs="sketchpad",
    outputs="label",
    title="Thai Digit Handwritten Classification",
    live=True
).launch(enable_queue=True)