[Python] How can I combine 5 trained models into a single one, and use the combined model for...

I have to implement federated learning. So I trained a list of 5 models, and trained each model on different datasets. Now I have to combine the trained models in the list but I have no Idea on how to do so. After that I have to use the combined model to predict on test set.

class MulticlassClassification:
def __init__(self, input_dims, layers, num_classes):
self.model = Sequential(name="server_model")
self.model.add(Dense(
layers[0],
input_shape=(input_dims,),
activation='relu'
))
for l in layers[1:]:
self.model.add(Dense(l, activation='relu'))
self.model.add(Dense(num_classes, activation='softmax'))
self.model.compile(
loss='sparse_categorical_crossentropy',
optimizer='adam',
metrics=['accuracy']
)
self.model.summary()
def set_avg_weights(self, models):
all_weights = np.array([
model.model.get_weights()
for model in models
])
avg_weights = np.mean(all_weights, axis=0)
self.model.set_weights(avg_weights)

all_weights = np.array([
model.model.get_weights()
for model in models
])
avg_weights = np.mean(all_weights, axis=0)
self.model.set_weights(avg_weights)
def fit(self, X, Y, epochs, batch_size):
self.model.fit(X, Y.to_numpy(), epochs=epochs, batch_size=batch_size)

def predict(self, X, Y):
preds = self.model.predict(X)
preds = [np.argmax(p) for p in preds]
print(classification_report(Y, preds))
ConfusionMatrixDisplay.from_predictions(Y, preds)


#Preprocessing and Training Simultaneously
label_encoder = preprocessing.LabelEncoder()
model = []
for i in range(5):
model.append(MulticlassClassification(27, [20, 20, 20], num_classes=5))
for i in range(5):
file_path = os.path.join(folder_path, nodes_list)
df = pd.read_csv(file_path)
df=df.drop(['FLAGS'],axis=1)
df['PKT_TYPE']=label_encoder.fit_transform(df['PKT_TYPE'])
df['PKT_CLASS']=label_encoder.fit_transform(df['PKT_CLASS'])
df['NODE_NAME_TO']=label_encoder.fit_transform(df['NODE_NAME_TO'])
df['NODE_NAME_FROM']=label_encoder.fit_transform(df['NODE_NAME_FROM'])
print(df['PKT_TYPE'].unique())
print(df['NODE_NAME_FROM'].unique())
print(df['NODE_NAME_TO'].unique())
print(df['PKT_CLASS'].unique())
X=df.drop(['PKT_CLASS'],axis=1)
Y=df['PKT_CLASS']
X1, X2, Y1, Y2= train_test_split(X, Y, train_size=0.9984375, random_state=1)
model.fit(X1, Y1, epochs=20, batch_size=32)

#Now I have to merge all models in the list model[] into a single one. How to do so??
#Then I have to predict the test set using the new model.```

Continue reading...