在训练前用一些权重实例化Keras模型

我有Keras模型：预先训练的CV模型+顶部添加的几层

我希望能够先做model.predict，再做model.fit

---

问：如何用一些权重（随机、零或其他）从截图中实例化模型？

这里是一个虚拟示例，用一些权重（随机、零或其他）初始化模型

def base_model(xx):
    
    x = Dense(32)(xx)
    x = Dense(8)(x)
    
    return Model(xx,x)

inp = Input((32,32,3))
x = base_model(inp)
x = GlobalAveragePooling2D()(x.output)
x = Dropout(0.3)(x)
out = Dense(10, activation='softmax')(x)

model = Model(inp,out)
model.summary()

# set weight with random number from a uniform... you can do the same also with zeros...
model.set_weights([np.random.uniform(0,1, i.shape) for i in model.get_weights()])