keras

one-hot encoding target
y = [1,2,3]
class 1 = [1,0,0]; class 2 = [0,1,0]; class 3 = [0,0,1]

from keras.utils import to_categorical
y=to_categorical(y)
y.shape

*******************************************************************
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.3,random_state=42)

********************************************************************

from sklearn.preprocessing import MinMaxScaler
scaler_object = MinMaxScaler()
scaler_object.fix(X_train)
scaled_X_train = scaler_object.transform(X_train)
scaled_X_test = scaler_object.transform(X_test)

********************************************************************
from keras.models import Sequential
from keras.layers import Dense

model = Sequential()
model.add(Dense(8,inout_dim=4,activation='relu'))
model.add(Dense(8,inout_dim=4,activation='relu'))
model.add(Dense(3,activation='softmax')) # 3 classes

model.compile(loss='catogorical_crossentropy', optimizer = 'adam', metrics=['accuracy'])

model.summary()

model.fit(scaled_X_train,y_train, epochs=150,verbose=2)

model.predict(scaled_X_test) # give probabilities
model.predict_classes(scaled_X_test) # give predicted class [1], [2] but we want [1,0,0], [0,1,0], cause problem when comparing with y_test

# transform y_test
y_test.argmax(axis=1)

from sklearn.metrics import confusion_matrix, classification_report, accuracy_score
confusion_matrix(y_test.argmax(axis=1), predictions))
accuracy_score(y_test.argmax(axis=1),predictions))

# save large model
model.save('modelname.h5')
from keras.models import load_model
new_model=load_model('modelname.h5')