def ___init___(self):
self.model = None
self.X_shape = None
self.Y_shape = None
self.classes = None
self.ckpt = None
Logger.success("DevnagiriCNN Model initialized")
def loadModel(self, height, width, channels, classes, path):
### Store input shape and classes
self.height = height
self.width = width
self.channels = channels
self.classes = classes
self.model = tf.keras.models.load_model(path)
Logger.success("Loaded pre-trained model at path: {}".format(path))
def train(self,
X,
Y,
epochs=5,
test_size=0.3,
batch_size=10,
model_save=False):
### Split data
x_train, x_test, y_train, y_test = prepareTrainData(
X,
Y,
self.height,
self.width,
self.channels,
self.classes,
test_size=test_size,
random_state=2)
### Create checkpoints
checkpoint1 = ModelCheckpoint(ckeckpoint_path,
verbose=1,
save_weights_only=True,
save_best_only=True,
mode='max')
callbacks_list = [checkpoint1]
# ### Train CNN model
self.model.fit(x_train,
y_train,
validation_data=(x_test, y_test),
epochs=epochs,
batch_size=batch_size,
callbacks=callbacks_list)
# ### Validate Trained model
val_loss, val_acc = self.model.evaluate(x_test, y_test, verbose=0)
Logger.info('Validation Results: val_loss {} val_acc {}'.format(
val_loss, val_acc))
# ### Save model
if (model_save == True):
self.model.save(model_path)
Logger.info("\nModel saved at {}".format(model_path))
elif (model_save != False):
try:
self.model.save(model_save)
Logger.success("Model saved at {}".format(model_save))
except Exception as e:
Logger.error(
"Error in filename or path given for saving model....")
self.model.save(model_path)
Logger.success("Model saved at {}".format(model_path))
else:
pass
def predict(self, X):
### Prepare predict data
x_pred = preparePredictData(X, self.height, self.width, self.channels)
### Predict with new inputs
predictions = self.model.predict(x_pred)
print("\n\n")
### Plotting first 10 inputs with predictions
for i in range(10):
predPlot(X[i], self.height, self.width)
prediction = np.argmax(predictions[i])
Logger.info("Prediction: {}".format(prediction))
return predictions
def validate(self, X, Y):
X_valid, Y_valid = prepareValidateData(X, Y, self.height, self.width,
self.channels, self.classes)
val_loss, val_acc = self.model.evaluate(X_valid, Y_valid)
Logger.info(
'Validation Results on new Test data : val_loss {} val_acc: {}'.
format(val_loss, val_acc))
### Predict with new inputs
val_predictions = self.model.predict(X_valid)
print("\n\n")
### Plotting first 10 inputs with predictions
for i in range(10):
predPlot(X[i], self.height, self.width)
val_prediction = np.argmax(val_predictions[i])
expected_output = Y_valid[i]
expected_corresponding_class = np.argmax(Y_valid[i])
Logger.info(
"Expected output: {} Prediction: {} Expected class: {}".
format(expected_output, val_prediction,
expected_corresponding_class))
def main():
Logger.init()
inp_option = Logger.menu()
try:
if inp_option == "q" or inp_option == "Q":
Logger.success("Exiting the application")
elif float(inp_option) == 1:
train()
elif float(inp_option) == 2:
test()
except Exception as e:
Logger.error(e)
def plot(X, Y, height, width):
z = X.reshape(height, width)
plt.imshow(z, cmap=plt.cm.binary)
plt.show()
Logger.info("Class label: {}".format(Y))