如何消除「UserWarning: The initializer GlorotUniform is unseeded」警告？

你提供的代码在运行bias_variance_decomp时会触发以下警告：

UserWarning: The initializer GlorotUniform is unseeded and being called multiple times, which will return identical values each time (even if the initializer is unseeded). Please update your code to provide a seed to the initializer, or avoid using the same initializer instance more than once.

原因是bias_variance_decomp会循环多次训练模型，但原代码仅创建了一个模型实例，默认的GlorotUniform初始化器未设置种子，重复调用会导致初始化值重复，触发TensorFlow的警告。

以下是两种可行的修改方案：

方案1：为初始化器显式设置种子 #

为每个Dense层的kernel_initializer和bias_initializer指定带独立种子的GlorotUniform实例，确保每次初始化生成不同的随机值：

import matplotlib.pyplot as plt
import numpy as np

import tensorflow as tf
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential
from tensorflow.keras.initializers import GlorotUniform

from mlxtend.evaluate import bias_variance_decomp
from mlxtend.data import boston_housing_data

from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import BaggingRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error

np.random.seed(16)
tf.random.set_seed(16)

X, y = boston_housing_data()
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                test_size=0.3,
                                                random_state=123,
                                                shuffle=True)

# 为每个层的初始化器设置独立种子
model = Sequential()
model.add(Dense(2048, activation='relu',
                kernel_initializer=GlorotUniform(seed=42),
                bias_initializer=GlorotUniform(seed=43)))
model.add(Dense(512, activation='relu',
                kernel_initializer=GlorotUniform(seed=44),
                bias_initializer=GlorotUniform(seed=45)))
model.add(Dense(32, activation='relu',
                kernel_initializer=GlorotUniform(seed=46),
                bias_initializer=GlorotUniform(seed=47)))
model.add(Dense(1, activation='linear',
                kernel_initializer=GlorotUniform(seed=48),
                bias_initializer=GlorotUniform(seed=49)))

optimizer = tf.keras.optimizers.Adam()
model.compile(loss='mean_squared_error', optimizer=optimizer)
model.fit(X_train, y_train, epochs=100, batch_size=32, verbose=0)
mean_squared_error(model.predict(X_test), y_test)

avg_expected_loss, avg_bias, avg_var = bias_variance_decomp(
    model, X_train, y_train, X_test, y_test, 
    loss='mse',
    num_rounds=100,
    random_seed=16,
    epochs=100,
    batch_size=32,
    verbose=0)

print('Average expected loss: %.3f' % avg_expected_loss)
print('Average bias: %.3f' % avg_bias)
print('Average variance: %.3f' % avg_var)

方案2：封装模型为函数（更推荐） #

偏差-方差分解的核心逻辑是每次训练独立的模型实例，将模型定义封装成函数，让bias_variance_decomp每次迭代都创建全新模型，从根源上避免重复使用初始化器：

import matplotlib.pyplot as plt
import numpy as np

import tensorflow as tf
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential

from mlxtend.evaluate import bias_variance_decomp
from mlxtend.data import boston_housing_data

from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import BaggingRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error

np.random.seed(16)
tf.random.set_seed(16)

X, y = boston_housing_data()
X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                test_size=0.3,
                                                random_state=123,
                                                shuffle=True)

# 将模型构建逻辑封装为函数
def build_model():
    model = Sequential()
    model.add(Dense(2048, activation='relu'))
    model.add(Dense(512, activation='relu'))
    model.add(Dense(32, activation='relu'))
    model.add(Dense(1, activation='linear'))
    optimizer = tf.keras.optimizers.Adam()
    model.compile(loss='mean_squared_error', optimizer=optimizer)
    return model

# 测试单个模型性能
model = build_model()
model.fit(X_train, y_train, epochs=100, batch_size=32, verbose=0)
mean_squared_error(model.predict(X_test), y_test)

# 传入模型构建函数而非实例
avg_expected_loss, avg_bias, avg_var = bias_variance_decomp(
    build_model, X_train, y_train, X_test, y_test, 
    loss='mse',
    num_rounds=100,
    random_seed=16,
    epochs=100,
    batch_size=32,
    verbose=0)

print('Average expected loss: %.3f' % avg_expected_loss)
print('Average bias: %.3f' % avg_bias)
print('Average variance: %.3f' % avg_var)

方案说明 #

方案2更贴合偏差-方差分解的设计逻辑，每次迭代都生成全新模型，既保证了初始化的随机性，又彻底消除警告；方案1通过设置种子避免重复初始化值，但本质仍是复用同一模型实例，规范性不如方案2。

内容的提问来源于stack exchange，提问作者Boris Reif