Add files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+saved_model.keras filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,13 +1,34 @@
----
-title: Image Digit Classification
-emoji: 📉
-colorFrom: yellow
-colorTo: yellow
-sdk: gradio
-sdk_version: 6.6.0
-app_file: app.py
-pinned: false
-short_description: This project is a machine learning application designed to r
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Digit Classification System
+This project is a machine learning application designed to recognize handwritten digits using Convolutional Neural 
+Networks (CNN). It includes tools for data preprocessing, model training, and an interactive drawing interface for real-time predictions.
+
+👉 [Live Demo](https://huggingface.co/spaces/Arcus72/Image-Digit-Classification)
+
+## Files
+- **app.py**: The main application file that creates a Gradio interface for the digit recognition system.
+- **main_analysis.ipynb**: A Jupyter notebook that contains the data processing, model training, and evaluation code.
+- **train_mnist.csv**: The dataset used for training the model.
+
+## Data source
+- **Source**: [MNIST in CSV (Kaggle)](https://www.kaggle.com/datasets/oddrationale/mnist-in-csv)
+- **Format**: CSV containing grayscale pixel values (0–255).
+- **Target**: The label column representing the digit (0–9).
+- **Dimensions**: Aside from the label, each row consists of 784 features (a flattened 28×28 pixel image).
+
+## Key Features
+- **Data processing for model training**: Processes raw CSV data into $28 \times 28$ grayscale images by parsing pixel 
+  coordinates (from train_mnist.csv).
+- **Dynamic Data Augmentation (new)**: Enhances model generalization using random rotations, translations, and zooms.
+- **Automated Hyperparameter Tuning (new)**: Utilizes BayesianOptimization to search for the best architecture, including the number of layers, filters, dropout rates, and optimizers.
+- **Comprehensive Evaluation**: Features detailed visualization of label distribution, confusion matrices, and an analysis of misclassified samples.
+- **Interactive Demo**: A Gradio interface that allows users to draw digits and get instant predictions from the trained model.
+
+## Tech Stack
+- **Core**: Python, TensorFlow/Keras
+- **Tuning**: Keras Tuner
+- **Data Science**: Pandas, NumPy, Scikit-learn
+- **Visualization**: Matplotlib, Tkinter, Pillow (PIL)
+- **UI**: Gradio
+
+## Results
+Based on the tuning process, the model typically achieves **>98%** accuracy on the test set.

app.py

@@ -0,0 +1,57 @@
+import os
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
+
+import numpy as np
+import gradio as gr
+from tensorflow.keras.models import load_model
+model = load_model('saved_model.keras')
+
+def predict_number(image):
+
+    image_rgb = image["composite"][..., :3] # delete alpha
+    image_grayScale= np.mean(image_rgb, axis=-1).astype(np.uint8)
+
+    image_grayScale = image_grayScale.reshape(1, 280, 280, 1)
+    print(image_grayScale)
+    preds = model.predict(image_grayScale)[0]
+
+    return {str(i): float(preds[i]) for i in range(10)}
+
+with gr.Blocks() as demo:
+    gr.Markdown("""
+    <center>
+    <h1>Digit Classification</h1>
+    </center>
+    
+    ##  How to use:
+    1. **Draw a single digit** (0-9) in the center of the black canvas.  Digit should be as large as possible.
+    2. **Click 'Recognize'** to see the prediction and the model's confidence levels.
+    3. **Click 'Clear canvas'** to clear the canvas.
+    
+    """)
+    sketch_pad = gr.Sketchpad(
+        label="Draw here (White on Black)",
+        layers=False,
+        canvas_size=(280, 280),
+        brush=gr.Brush(colors=["#FFFFFF"], default_color="#FFFFFF", default_size=10),
+        value={
+            "background": np.zeros((280, 280, 3), dtype=np.uint8),
+            "layers": [],
+            "composite": None
+        },
+        type="numpy"
+    )
+
+    btn = gr.Button("Recognize", variant="primary")
+    clear_btn = gr.Button("Clear canvas", variant="stop")
+    output = gr.Label(num_top_classes=3, label="Prediction")
+
+    clear_btn.click(lambda: {"background": np.zeros((280, 280, 3), dtype=np.uint8), "layers": [], "composite": None},
+                    outputs=sketch_pad)
+    btn.click(predict_number, inputs=sketch_pad, outputs=output)
+    gr.Markdown("""
+    ## GitHub Repository
+    If you want to see the training process, data augmentation, and how the model was built,  
+        check out the **[main_analysis.ipynb](https://github.com/Arcus72/Image-Digit-Classification)** file in my GitHub repository.""")
+
+demo.launch(theme=gr.themes.Soft())

saved_model.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a0b7b9c0fb635c4dd8d0cbcbcb209cef9b17eaa9bba3579586828482d94a0ba
+size 6964085