Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -34,28 +34,27 @@ def create_model():
|
|
34 |
model.compile(optimizer="adam", loss="categorical_crossentropy", metrics=["accuracy"])
|
35 |
return model
|
36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
# Custom callback for logging
|
38 |
class StreamlitLogger(keras.callbacks.Callback):
|
39 |
def on_epoch_end(self, epoch, logs=None):
|
40 |
-
if logs is
|
41 |
logs = {}
|
42 |
-
|
43 |
st.write(f"Epoch {epoch + 1}:")
|
44 |
st.write(f" Train Loss: {logs.get('loss'):.4f}")
|
45 |
st.write(f" Train Accuracy: {logs.get('accuracy'):.4f}")
|
46 |
st.write(f" Val Loss: {logs.get('val_loss'):.4f}")
|
47 |
st.write(f" Val Accuracy: {logs.get('val_accuracy'):.4f}")
|
48 |
|
49 |
-
# Streamlit UI
|
50 |
-
st.title("CNN for MNIST Classification")
|
51 |
-
|
52 |
-
# Check if model is saved
|
53 |
-
model_path = "mnist_cnn_model.h5"
|
54 |
-
|
55 |
if st.button("Train Model"):
|
56 |
model = create_model()
|
57 |
|
58 |
-
# Create logger instance
|
59 |
logger = StreamlitLogger()
|
60 |
|
61 |
with st.spinner("Training..."):
|
@@ -72,9 +71,44 @@ if st.button("Train Model"):
|
|
72 |
ax1.set_title("Training and Validation Loss")
|
73 |
ax1.set_xlabel("Epoch")
|
74 |
ax1.set_ylabel("Loss")
|
75 |
-
ax1.legend()
|
76 |
|
77 |
ax2.plot(history.history["accuracy"], label="Train Accuracy")
|
78 |
ax2.plot(history.history["val_accuracy"], label="Val Accuracy")
|
79 |
ax2.set_title("Training and Validation Accuracy")
|
80 |
-
ax2.set_xlabel
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
model.compile(optimizer="adam", loss="categorical_crossentropy", metrics=["accuracy"])
|
35 |
return model
|
36 |
|
37 |
+
# Streamlit UI
|
38 |
+
st.title("CNN for MNIST Classification")
|
39 |
+
|
40 |
+
# Check if model is saved
|
41 |
+
model_path = "mnist_cnn_model.h5"
|
42 |
+
|
43 |
# Custom callback for logging
|
44 |
class StreamlitLogger(keras.callbacks.Callback):
|
45 |
def on_epoch_end(self, epoch, logs=None):
|
46 |
+
if logs is none:
|
47 |
logs = {}
|
48 |
+
|
49 |
st.write(f"Epoch {epoch + 1}:")
|
50 |
st.write(f" Train Loss: {logs.get('loss'):.4f}")
|
51 |
st.write(f" Train Accuracy: {logs.get('accuracy'):.4f}")
|
52 |
st.write(f" Val Loss: {logs.get('val_loss'):.4f}")
|
53 |
st.write(f" Val Accuracy: {logs.get('val_accuracy'):.4f}")
|
54 |
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
if st.button("Train Model"):
|
56 |
model = create_model()
|
57 |
|
|
|
58 |
logger = StreamlitLogger()
|
59 |
|
60 |
with st.spinner("Training..."):
|
|
|
71 |
ax1.set_title("Training and Validation Loss")
|
72 |
ax1.set_xlabel("Epoch")
|
73 |
ax1.set_ylabel("Loss")
|
|
|
74 |
|
75 |
ax2.plot(history.history["accuracy"], label="Train Accuracy")
|
76 |
ax2.plot(history.history["val_accuracy"], label="Val Accuracy")
|
77 |
ax2.set_title("Training and Validation Accuracy")
|
78 |
+
ax2.set_xlabel("Epoch")
|
79 |
+
ax2.set_ylabel("Accuracy")
|
80 |
+
|
81 |
+
ax1.legend()
|
82 |
+
ax2.legend()
|
83 |
+
|
84 |
+
st.pyplot(fig)
|
85 |
+
|
86 |
+
# Evaluate the model on test data
|
87 |
+
test_preds = np.argmax(model.predict(test_images), axis=1)
|
88 |
+
true_labels = np.argmax(test_labels, axis=1)
|
89 |
+
|
90 |
+
st.session_state['true_labels'] = true_labels
|
91 |
+
|
92 |
+
report = classification_report(true_labels, test_preds, digits=4)
|
93 |
+
st.text("Classification Report:")
|
94 |
+
st.text(report)
|
95 |
+
|
96 |
+
index = st.number_input("Enter an index (0-9999) to test:", min_value=0, max_value=9999, step=1)
|
97 |
+
|
98 |
+
def test_index_prediction(index):
|
99 |
+
image = test_images[index].reshape(28, 28)
|
100 |
+
st.image(image, caption=f"True Label: {st.session_state['true_labels'][index]}", use_column_width=True)
|
101 |
+
|
102 |
+
# Reload the model
|
103 |
+
if not os.path.exists(model_path):
|
104 |
+
st.error("Train the model first.")
|
105 |
+
return
|
106 |
+
|
107 |
+
model = keras.models.load_model(model_path)
|
108 |
+
|
109 |
+
prediction = model.predict(test_images[index].reshape(1, 28, 28, 1))
|
110 |
+
predicted_class = np.argmax(prediction)
|
111 |
+
st.write(f"Predicted Class: {predicted_class}")
|
112 |
+
|
113 |
+
if st.button("Test Index"):
|
114 |
+
test_index_prediction(index)
|