GPT-4 Generated readable code, but there are minor problems when the prompt becomes complicated. In the first example code for generating the deepFM, there was a bug after running directly the code generated by GPT-4. The root cause is that it used the wrong indices to extract the categorical features.
Note: The prompt is more complicated and will be added later. The following code is the right code after debugging and fixing the bug generated by GPT-4
import numpy as np
import pandas as pd
import tensorflow as tf
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from tensorflow.keras.layers import (LSTM, Add, Concatenate, Dense, Embedding,
Flatten, Input, Lambda, Multiply)
from tensorflow.keras.models import Model
# 示例数据，你需要用你自己的数据替换
data = {
'gender': ['M', 'F', 'M', 'F'],
'age': [25, 30, 35, 28],
'user_behavior': [[1, 2, 1, 3], [1, 1, 3], [1, 2, 2, 1], [3, 1, 2]],
'CVR': [0, 1, 0, 1]
}
df = pd.DataFrame(data)
# 数值特征
numerical_features = ['age']
numerical_data = df[numerical_features].values
# 标准化数值特征
scaler = StandardScaler()
numerical_data = scaler.fit_transform(numerical_data)
# 类别特征
categorical_features = ['gender']
categorical_data = df[categorical_features].values
# Count the number of unique values for each categorical feature
num_unique_values = [len(df[feature].unique())
for feature in categorical_features]
# Calculate the total number of unique categorical feature values
num_categorical_features = sum(num_unique_values)
# 独热编码类别特征
encoder = OneHotEncoder()
categorical_data = encoder.fit_transform(categorical_data).toarray()
print(categorical_data)
# 处理用户行为数据
max_length = max([len(x) for x in df['user_behavior']])
user_behavior_padded = tf.keras.preprocessing.sequence.pad_sequences(
df['user_behavior'], maxlen=max_length)
# 合并所有特征
X = np.hstack([numerical_data, categorical_data])
# 特征和标签分离
y = df['CVR'].values
# 数据划分为训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42)
# LSTM 输入处理
X_train_behavior = user_behavior_padded[:len(X_train)]
X_test_behavior = user_behavior_padded[len(X_train):]
# 其他特征输入
X_train_features = X_train_features = X_train
X_test_features = X_test
# 创建 LSTM 模型
lstm_input = Input(shape=(max_length,))
lstm_output = Embedding(max_length, 8)(lstm_input)
lstm_output = LSTM(16)(lstm_output)
# DeepFM 模型
features_input = Input(shape=(X_train_features.shape[1],))
# 线性部分
linear_output = Dense(1, activation='linear')(features_input)
# FM 部分
embedding_layer = Embedding(num_categorical_features, 8)
embeddings = []
for i in range(num_categorical_features):
emb = embedding_layer(features_input[:, i+len(numerical_features)])
embeddings.append(emb)
embeddings = tf.stack(embeddings, axis=1)
fm_sum_of_squares = tf.square(tf.reduce_sum(embeddings, axis=1))
fm_square_of_sum = tf.reduce_sum(tf.square(embeddings), axis=1)
fm_output = 0.5 * (fm_sum_of_squares - fm_square_of_sum)
# DNN 部分
deep_output = Dense(64, activation='relu')(features_input)
deep_output = Dense(32, activation='relu')(deep_output)
# 合并 DeepFM 输出
# deepfm_output = Add()([linear_output, fm_output, deep_output])
deepfm_output = Concatenate()([linear_output, fm_output, deep_output])
# 合并 DeepFM 输出和 LSTM 输出
combined_output = Concatenate()([lstm_output, deepfm_output])
combined_output = Dense(32, activation='relu')(combined_output)
output = Dense(1, activation='sigmoid')(combined_output)
# 创建模型
model = Model([lstm_input, features_input], output)
# 编译模型
model.compile(optimizer='adam', loss='binary_crossentropy')
config = model.get_config() # Returns pretty much every information about your model
print(config["layers"]) # returns a tuple of width, height and channels
print(len([X_train_behavior, X_train_features]))
print(X_train_behavior.shape)
print(X_train_features.shape)
# 训练模型
model.fit([X_train_behavior, X_train_features],
y_train, epochs=3, batch_size=1, verbose=1)
# 预测
y_pred = model.predict([X_test_behavior, X_test_features])
print("预测结果：", y_pred)
[[0. 1.]
[1. 0.]
[0. 1.]
[1. 0.]]
[{'class_name': 'InputLayer', 'config': {'batch_input_shape': (None, 3), 'dtype': 'float32', 'sparse': False, 'ragged': False, 'name': 'input_2'}, 'name': 'input_2', 'inbound_nodes': []}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem', 'inbound_nodes': [['input_2', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 0)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_1', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_1', 'inbound_nodes': [['input_2', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 1)}]]}, {'class_name': 'Embedding', 'config': {'name': 'embedding_1', 'trainable': True, 'dtype': 'float32', 'batch_input_shape': (None, None), 'input_dim': 2, 'output_dim': 8, 'embeddings_initializer': {'class_name': 'RandomUniform', 'config': {'minval': -0.05, 'maxval': 0.05, 'seed': None}}, 'embeddings_regularizer': None, 'activity_regularizer': None, 'embeddings_constraint': None, 'mask_zero': False, 'input_length': None}, 'name': 'embedding_1', 'inbound_nodes': [[['tf.__operators__.getitem', 0, 0, {}]], [['tf.__operators__.getitem_1', 0, 0, {}]]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.stack', 'trainable': True, 'dtype': 'float32', 'function': 'stack'}, 'name': 'tf.stack', 'inbound_nodes': [[['embedding_1', 0, 0, {'axis': 1}], ['embedding_1', 1, 0, {'axis': 1}]]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.reduce_sum', 'trainable': True, 'dtype': 'float32', 'function': 'math.reduce_sum'}, 'name': 'tf.math.reduce_sum', 'inbound_nodes': [['tf.stack', 0, 0, {'axis': 1}]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.square_1', 'trainable': True, 'dtype': 'float32', 'function': 'math.square'}, 'name': 'tf.math.square_1', 'inbound_nodes': [['tf.stack', 0, 0, {'name': None}]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.square', 'trainable': True, 'dtype': 'float32', 'function': 'math.square'}, 'name': 'tf.math.square', 'inbound_nodes': [['tf.math.reduce_sum', 0, 0, {'name': None}]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.reduce_sum_1', 'trainable': True, 'dtype': 'float32', 'function': 'math.reduce_sum'}, 'name': 'tf.math.reduce_sum_1', 'inbound_nodes': [['tf.math.square_1', 0, 0, {'axis': 1}]]}, {'class_name': 'InputLayer', 'config': {'batch_input_shape': (None, 4), 'dtype': 'float32', 'sparse': False, 'ragged': False, 'name': 'input_1'}, 'name': 'input_1', 'inbound_nodes': []}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.subtract', 'trainable': True, 'dtype': 'float32', 'function': 'math.subtract'}, 'name': 'tf.math.subtract', 'inbound_nodes': [['tf.math.square', 0, 0, {'y': ['tf.math.reduce_sum_1', 0, 0], 'name': None}]]}, {'class_name': 'Dense', 'config': {'name': 'dense_1', 'trainable': True, 'dtype': 'float32', 'units': 64, 'activation': 'relu', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_1', 'inbound_nodes': [[['input_2', 0, 0, {}]]]}, {'class_name': 'Embedding', 'config': {'name': 'embedding', 'trainable': True, 'dtype': 'float32', 'batch_input_shape': (None, None), 'input_dim': 4, 'output_dim': 8, 'embeddings_initializer': {'class_name': 'RandomUniform', 'config': {'minval': -0.05, 'maxval': 0.05, 'seed': None}}, 'embeddings_regularizer': None, 'activity_regularizer': None, 'embeddings_constraint': None, 'mask_zero': False, 'input_length': None}, 'name': 'embedding', 'inbound_nodes': [[['input_1', 0, 0, {}]]]}, {'class_name': 'Dense', 'config': {'name': 'dense', 'trainable': True, 'dtype': 'float32', 'units': 1, 'activation': 'linear', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense', 'inbound_nodes': [[['input_2', 0, 0, {}]]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.multiply', 'trainable': True, 'dtype': 'float32', 'function': 'math.multiply'}, 'name': 'tf.math.multiply', 'inbound_nodes': [['_CONSTANT_VALUE', -1, 0.5, {'y': ['tf.math.subtract', 0, 0], 'name': None}]]}, {'class_name': 'Dense', 'config': {'name': 'dense_2', 'trainable': True, 'dtype': 'float32', 'units': 32, 'activation': 'relu', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_2', 'inbound_nodes': [[['dense_1', 0, 0, {}]]]}, {'class_name': 'LSTM', 'config': {'name': 'lstm', 'trainable': True, 'dtype': 'float32', 'return_sequences': False, 'return_state': False, 'go_backwards': False, 'stateful': False, 'unroll': False, 'time_major': False, 'units': 16, 'activation': 'tanh', 'recurrent_activation': 'sigmoid', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}, 'shared_object_id': 24}, 'recurrent_initializer': {'class_name': 'Orthogonal', 'config': {'gain': 1.0, 'seed': None}, 'shared_object_id': 25}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}, 'shared_object_id': 26}, 'unit_forget_bias': True, 'kernel_regularizer': None, 'recurrent_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'recurrent_constraint': None, 'bias_constraint': None, 'dropout': 0.0, 'recurrent_dropout': 0.0, 'implementation': 2}, 'name': 'lstm', 'inbound_nodes': [[['embedding', 0, 0, {}]]]}, {'class_name': 'Concatenate', 'config': {'name': 'concatenate', 'trainable': True, 'dtype': 'float32', 'axis': -1}, 'name': 'concatenate', 'inbound_nodes': [[['dense', 0, 0, {}], ['tf.math.multiply', 0, 0, {}], ['dense_2', 0, 0, {}]]]}, {'class_name': 'Concatenate', 'config': {'name': 'concatenate_1', 'trainable': True, 'dtype': 'float32', 'axis': -1}, 'name': 'concatenate_1', 'inbound_nodes': [[['lstm', 0, 0, {}], ['concatenate', 0, 0, {}]]]}, {'class_name': 'Dense', 'config': {'name': 'dense_3', 'trainable': True, 'dtype': 'float32', 'units': 32, 'activation': 'relu', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_3', 'inbound_nodes': [[['concatenate_1', 0, 0, {}]]]}, {'class_name': 'Dense', 'config': {'name': 'dense_4', 'trainable': True, 'dtype': 'float32', 'units': 1, 'activation': 'sigmoid', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_4', 'inbound_nodes': [[['dense_3', 0, 0, {}]]]}]
2
(3, 4)
(3, 3)
Epoch 1/3
3/3 [==============================] - 16s 22ms/step - loss: 0.6742
Epoch 2/3
3/3 [==============================] - 0s 19ms/step - loss: 0.6386
Epoch 3/3
3/3 [==============================] - 0s 13ms/step - loss: 0.6085
1/1 [==============================] - 5s 5s/step
预测结果： [[0.52413774]]
data = {
'gender': ['M', 'F', 'M', 'F'],
'age': [25, 30, 35, 28],
'location': ['A', 'B', 'A', 'B'],
'interests': ['sports', 'fashion', 'technology', 'music'],
'device': ['mobile', 'tablet', 'mobile', 'desktop'],
'purchase_history': [0, 1, 0, 1],
'ad_category': ['electronics', 'fashion', 'fashion', 'electronics'],
'advertiser': ['X', 'Y', 'Y', 'X'],
'ad_type': ['image', 'video', 'carousel', 'image'],
'ad_channel': ['google', 'facebook', 'twitter', 'google'],
'ad_time': [8, 15, 21, 10],
'page_type': ['blog', 'shopping', 'news', 'blog'],
'page_load_speed': [2.5, 3.2, 1.8, 2.0],
'page_layout': ['A', 'B', 'C', 'A'],
'user_behavior': [[1, 2, 1, 3], [1, 1, 3], [1, 2, 2, 1], [3, 1, 2]],
'CVR': [0, 1, 0, 1]
}
Example Output:
[[0. 1. 1. 0. 0. 0. 1. 0. 0. 1. 0. 1. 0. 1. 0. 0. 1. 0. 0. 1. 0. 1. 0. 0.
1. 0. 0.]
[1. 0. 0. 1. 1. 0. 0. 0. 0. 0. 1. 0. 1. 0. 1. 0. 0. 1. 1. 0. 0. 0. 0. 1.
0. 1. 0.]
[0. 1. 1. 0. 0. 0. 0. 1. 0. 1. 0. 0. 1. 0. 1. 1. 0. 0. 0. 0. 1. 0. 1. 0.
0. 0. 1.]
[1. 0. 0. 1. 0. 1. 0. 0. 1. 0. 0. 1. 0. 1. 0. 0. 1. 0. 0. 1. 0. 1. 0. 0.
1. 0. 0.]]
[{'class_name': 'InputLayer', 'config': {'batch_input_shape': (None, 31), 'dtype': 'float32', 'sparse': False, 'ragged': False, 'name': 'input_12'}, 'name': 'input_12', 'inbound_nodes': []}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_9', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_9', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 4)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_10', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_10', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 5)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_11', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_11', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 6)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_12', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_12', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 7)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_13', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_13', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 8)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_14', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_14', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 9)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_15', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_15', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 10)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_16', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_16', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 11)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_17', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_17', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 12)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_18', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_18', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 13)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_19', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_19', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 14)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_20', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_20', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 15)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_21', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_21', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 16)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_22', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_22', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 17)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_23', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_23', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 18)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_24', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_24', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 19)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_25', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_25', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 20)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_26', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_26', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 21)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_27', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_27', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 22)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_28', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_28', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 23)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_29', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_29', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 24)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_30', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_30', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 25)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_31', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_31', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 26)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_32', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_32', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 27)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_33', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_33', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 28)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_34', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_34', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 29)}]]}, {'class_name': 'SlicingOpLambda', 'config': {'name': 'tf.__operators__.getitem_35', 'trainable': True, 'dtype': 'float32', 'function': '__operators__.getitem'}, 'name': 'tf.__operators__.getitem_35', 'inbound_nodes': [['input_12', 0, 0, {'slice_spec': ({'start': None, 'stop': None, 'step': None}, 30)}]]}, {'class_name': 'Embedding', 'config': {'name': 'embedding_11', 'trainable': True, 'dtype': 'float32', 'batch_input_shape': (None, None), 'input_dim': 27, 'output_dim': 8, 'embeddings_initializer': {'class_name': 'RandomUniform', 'config': {'minval': -0.05, 'maxval': 0.05, 'seed': None}}, 'embeddings_regularizer': None, 'activity_regularizer': None, 'embeddings_constraint': None, 'mask_zero': False, 'input_length': None}, 'name': 'embedding_11', 'inbound_nodes': [[['tf.__operators__.getitem_9', 0, 0, {}]], [['tf.__operators__.getitem_10', 0, 0, {}]], [['tf.__operators__.getitem_11', 0, 0, {}]], [['tf.__operators__.getitem_12', 0, 0, {}]], [['tf.__operators__.getitem_13', 0, 0, {}]], [['tf.__operators__.getitem_14', 0, 0, {}]], [['tf.__operators__.getitem_15', 0, 0, {}]], [['tf.__operators__.getitem_16', 0, 0, {}]], [['tf.__operators__.getitem_17', 0, 0, {}]], [['tf.__operators__.getitem_18', 0, 0, {}]], [['tf.__operators__.getitem_19', 0, 0, {}]], [['tf.__operators__.getitem_20', 0, 0, {}]], [['tf.__operators__.getitem_21', 0, 0, {}]], [['tf.__operators__.getitem_22', 0, 0, {}]], [['tf.__operators__.getitem_23', 0, 0, {}]], [['tf.__operators__.getitem_24', 0, 0, {}]], [['tf.__operators__.getitem_25', 0, 0, {}]], [['tf.__operators__.getitem_26', 0, 0, {}]], [['tf.__operators__.getitem_27', 0, 0, {}]], [['tf.__operators__.getitem_28', 0, 0, {}]], [['tf.__operators__.getitem_29', 0, 0, {}]], [['tf.__operators__.getitem_30', 0, 0, {}]], [['tf.__operators__.getitem_31', 0, 0, {}]], [['tf.__operators__.getitem_32', 0, 0, {}]], [['tf.__operators__.getitem_33', 0, 0, {}]], [['tf.__operators__.getitem_34', 0, 0, {}]], [['tf.__operators__.getitem_35', 0, 0, {}]]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.stack_4', 'trainable': True, 'dtype': 'float32', 'function': 'stack'}, 'name': 'tf.stack_4', 'inbound_nodes': [[['embedding_11', 0, 0, {'axis': 1}], ['embedding_11', 1, 0, {'axis': 1}], ['embedding_11', 2, 0, {'axis': 1}], ['embedding_11', 3, 0, {'axis': 1}], ['embedding_11', 4, 0, {'axis': 1}], ['embedding_11', 5, 0, {'axis': 1}], ['embedding_11', 6, 0, {'axis': 1}], ['embedding_11', 7, 0, {'axis': 1}], ['embedding_11', 8, 0, {'axis': 1}], ['embedding_11', 9, 0, {'axis': 1}], ['embedding_11', 10, 0, {'axis': 1}], ['embedding_11', 11, 0, {'axis': 1}], ['embedding_11', 12, 0, {'axis': 1}], ['embedding_11', 13, 0, {'axis': 1}], ['embedding_11', 14, 0, {'axis': 1}], ['embedding_11', 15, 0, {'axis': 1}], ['embedding_11', 16, 0, {'axis': 1}], ['embedding_11', 17, 0, {'axis': 1}], ['embedding_11', 18, 0, {'axis': 1}], ['embedding_11', 19, 0, {'axis': 1}], ['embedding_11', 20, 0, {'axis': 1}], ['embedding_11', 21, 0, {'axis': 1}], ['embedding_11', 22, 0, {'axis': 1}], ['embedding_11', 23, 0, {'axis': 1}], ['embedding_11', 24, 0, {'axis': 1}], ['embedding_11', 25, 0, {'axis': 1}], ['embedding_11', 26, 0, {'axis': 1}]]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.reduce_sum_8', 'trainable': True, 'dtype': 'float32', 'function': 'math.reduce_sum'}, 'name': 'tf.math.reduce_sum_8', 'inbound_nodes': [['tf.stack_4', 0, 0, {'axis': 1}]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.square_9', 'trainable': True, 'dtype': 'float32', 'function': 'math.square'}, 'name': 'tf.math.square_9', 'inbound_nodes': [['tf.stack_4', 0, 0, {'name': None}]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.square_8', 'trainable': True, 'dtype': 'float32', 'function': 'math.square'}, 'name': 'tf.math.square_8', 'inbound_nodes': [['tf.math.reduce_sum_8', 0, 0, {'name': None}]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.reduce_sum_9', 'trainable': True, 'dtype': 'float32', 'function': 'math.reduce_sum'}, 'name': 'tf.math.reduce_sum_9', 'inbound_nodes': [['tf.math.square_9', 0, 0, {'axis': 1}]]}, {'class_name': 'InputLayer', 'config': {'batch_input_shape': (None, 4), 'dtype': 'float32', 'sparse': False, 'ragged': False, 'name': 'input_11'}, 'name': 'input_11', 'inbound_nodes': []}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.subtract_4', 'trainable': True, 'dtype': 'float32', 'function': 'math.subtract'}, 'name': 'tf.math.subtract_4', 'inbound_nodes': [['tf.math.square_8', 0, 0, {'y': ['tf.math.reduce_sum_9', 0, 0], 'name': None}]]}, {'class_name': 'Dense', 'config': {'name': 'dense_22', 'trainable': True, 'dtype': 'float32', 'units': 64, 'activation': 'relu', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_22', 'inbound_nodes': [[['input_12', 0, 0, {}]]]}, {'class_name': 'Embedding', 'config': {'name': 'embedding_10', 'trainable': True, 'dtype': 'float32', 'batch_input_shape': (None, None), 'input_dim': 4, 'output_dim': 8, 'embeddings_initializer': {'class_name': 'RandomUniform', 'config': {'minval': -0.05, 'maxval': 0.05, 'seed': None}}, 'embeddings_regularizer': None, 'activity_regularizer': None, 'embeddings_constraint': None, 'mask_zero': False, 'input_length': None}, 'name': 'embedding_10', 'inbound_nodes': [[['input_11', 0, 0, {}]]]}, {'class_name': 'Dense', 'config': {'name': 'dense_21', 'trainable': True, 'dtype': 'float32', 'units': 1, 'activation': 'linear', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_21', 'inbound_nodes': [[['input_12', 0, 0, {}]]]}, {'class_name': 'TFOpLambda', 'config': {'name': 'tf.math.multiply_4', 'trainable': True, 'dtype': 'float32', 'function': 'math.multiply'}, 'name': 'tf.math.multiply_4', 'inbound_nodes': [['_CONSTANT_VALUE', -1, 0.5, {'y': ['tf.math.subtract_4', 0, 0], 'name': None}]]}, {'class_name': 'Dense', 'config': {'name': 'dense_23', 'trainable': True, 'dtype': 'float32', 'units': 32, 'activation': 'relu', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_23', 'inbound_nodes': [[['dense_22', 0, 0, {}]]]}, {'class_name': 'LSTM', 'config': {'name': 'lstm_5', 'trainable': True, 'dtype': 'float32', 'return_sequences': False, 'return_state': False, 'go_backwards': False, 'stateful': False, 'unroll': False, 'time_major': False, 'units': 16, 'activation': 'tanh', 'recurrent_activation': 'sigmoid', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}, 'shared_object_id': 49}, 'recurrent_initializer': {'class_name': 'Orthogonal', 'config': {'gain': 1.0, 'seed': None}, 'shared_object_id': 50}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}, 'shared_object_id': 51}, 'unit_forget_bias': True, 'kernel_regularizer': None, 'recurrent_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'recurrent_constraint': None, 'bias_constraint': None, 'dropout': 0.0, 'recurrent_dropout': 0.0, 'implementation': 2}, 'name': 'lstm_5', 'inbound_nodes': [[['embedding_10', 0, 0, {}]]]}, {'class_name': 'Concatenate', 'config': {'name': 'concatenate_8', 'trainable': True, 'dtype': 'float32', 'axis': -1}, 'name': 'concatenate_8', 'inbound_nodes': [[['dense_21', 0, 0, {}], ['tf.math.multiply_4', 0, 0, {}], ['dense_23', 0, 0, {}]]]}, {'class_name': 'Concatenate', 'config': {'name': 'concatenate_9', 'trainable': True, 'dtype': 'float32', 'axis': -1}, 'name': 'concatenate_9', 'inbound_nodes': [[['lstm_5', 0, 0, {}], ['concatenate_8', 0, 0, {}]]]}, {'class_name': 'Dense', 'config': {'name': 'dense_24', 'trainable': True, 'dtype': 'float32', 'units': 32, 'activation': 'relu', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_24', 'inbound_nodes': [[['concatenate_9', 0, 0, {}]]]}, {'class_name': 'Dense', 'config': {'name': 'dense_25', 'trainable': True, 'dtype': 'float32', 'units': 1, 'activation': 'sigmoid', 'use_bias': True, 'kernel_initializer': {'class_name': 'GlorotUniform', 'config': {'seed': None}}, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'kernel_regularizer': None, 'bias_regularizer': None, 'activity_regularizer': None, 'kernel_constraint': None, 'bias_constraint': None}, 'name': 'dense_25', 'inbound_nodes': [[['dense_24', 0, 0, {}]]]}]
2
(3, 4)
(3, 31)
Epoch 1/3
3/3 [==============================] - 3s 7ms/step - loss: 0.6556
Epoch 2/3
3/3 [==============================] - 0s 12ms/step - loss: 0.5769
Epoch 3/3
3/3 [==============================] - 0s 12ms/step - loss: 0.5104
WARNING:tensorflow:5 out of the last 5 calls to <function Model.make_predict_function.<locals>.predict_function at 0x7efc39194a60> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has reduce_retracing=True option that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for more details.
1/1 [==============================] - 1s 778ms/step
预测结果： [[0.29611892]]
Wrong Code:
# 合并所有特征
X = np.hstack([numerical_data, categorical_data])
# FM 部分
embedding_layer = Embedding(num_categorical_features, 8)
embeddings = []
for i in range(num_categorical_features):
emb = embedding_layer(features_input[:, i])
embeddings.append(emb)
Right Code:
# 合并所有特征
X = np.hstack([numerical_data, categorical_data])
# FM 部分
embedding_layer = Embedding(num_categorical_features, 8)
embeddings = []
for i in range(num_categorical_features):
emb = embedding_layer(features_input[:, i+len(numerical_features)])
embeddings.append(emb)
The root cause is that the following code X = np.hstack([numerical_data, categorical_data]) merges numerical features and categorical features together. However the FM part is only supposed to process the categorical features, and it should not start with fetching numerical features. Thus, we need to change from emb = embedding_layer(features_input[:, i]) to emb = embedding_layer(features_input[:, i+len(numerical_features)]).
Without machine learning debugging experience, it will be hard to debug in this case. By only using GPT-4 prompt, GPT-4 cannot find out the root cause.
The root cause can be found in the following way with machine learning domain knowledge and debugger, the error trace said that indices[0] = -1 is not in [0, 2), and the categorical features are with 2 dimensions, then we can narrow down the problem to the embeddings for categorical features. In addition, normally the numercial features would have negative values like -1, while in this case categorical features are with one-hot encoding and the values of categorical features should not be zero.
/usr/local/lib/python3.9/dist-packages/tensorflow/python/eager/execute.py in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
50 try:
51 ctx.ensure_initialized()
---> 52 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
53 inputs, attrs, num_outputs)
54 except core._NotOkStatusException as e:
InvalidArgumentError: Graph execution error:
Detected at node 'model_20/embedding_39/embedding_lookup_1' defined at (most recent call last):
File "/usr/lib/python3.9/runpy.py", line 197, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/usr/lib/python3.9/runpy.py", line 87, in _run_code
exec(code, run_globals)
File "/usr/local/lib/python3.9/dist-packages/ipykernel_launcher.py", line 16, in <module>
app.launch_new_instance()
File "/usr/local/lib/python3.9/dist-packages/traitlets/config/application.py", line 992, in launch_instance
app.start()
File "/usr/local/lib/python3.9/dist-packages/ipykernel/kernelapp.py", line 619, in start
self.io_loop.start()
File "/usr/local/lib/python3.9/dist-packages/tornado/platform/asyncio.py", line 215, in start
self.asyncio_loop.run_forever()
File "/usr/lib/python3.9/asyncio/base_events.py", line 601, in run_forever
self._run_once()
File "/usr/lib/python3.9/asyncio/base_events.py", line 1905, in _run_once
handle._run()
File "/usr/lib/python3.9/asyncio/events.py", line 80, in _run
self._context.run(self._callback, *self._args)
File "/usr/local/lib/python3.9/dist-packages/tornado/ioloop.py", line 687, in <lambda>
lambda f: self._run_callback(functools.partial(callback, future))
File "/usr/local/lib/python3.9/dist-packages/tornado/ioloop.py", line 740, in _run_callback
ret = callback()
File "/usr/local/lib/python3.9/dist-packages/tornado/gen.py", line 821, in inner
self.ctx_run(self.run)
File "/usr/local/lib/python3.9/dist-packages/tornado/gen.py", line 782, in run
yielded = self.gen.send(value)
File "/usr/local/lib/python3.9/dist-packages/ipykernel/kernelbase.py", line 361, in process_one
yield gen.maybe_future(dispatch(*args))
File "/usr/local/lib/python3.9/dist-packages/tornado/gen.py", line 234, in wrapper
yielded = ctx_run(next, result)
File "/usr/local/lib/python3.9/dist-packages/ipykernel/kernelbase.py", line 261, in dispatch_shell
yield gen.maybe_future(handler(stream, idents, msg))
File "/usr/local/lib/python3.9/dist-packages/tornado/gen.py", line 234, in wrapper
yielded = ctx_run(next, result)
File "/usr/local/lib/python3.9/dist-packages/ipykernel/kernelbase.py", line 539, in execute_request
self.do_execute(
File "/usr/local/lib/python3.9/dist-packages/tornado/gen.py", line 234, in wrapper
yielded = ctx_run(next, result)
File "/usr/local/lib/python3.9/dist-packages/ipykernel/ipkernel.py", line 302, in do_execute
res = shell.run_cell(code, store_history=store_history, silent=silent)
File "/usr/local/lib/python3.9/dist-packages/ipykernel/zmqshell.py", line 539, in run_cell
return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/IPython/core/interactiveshell.py", line 2975, in run_cell
result = self._run_cell(
File "/usr/local/lib/python3.9/dist-packages/IPython/core/interactiveshell.py", line 3030, in _run_cell
return runner(coro)
File "/usr/local/lib/python3.9/dist-packages/IPython/core/async_helpers.py", line 78, in _pseudo_sync_runner
coro.send(None)
File "/usr/local/lib/python3.9/dist-packages/IPython/core/interactiveshell.py", line 3257, in run_cell_async
has_raised = await self.run_ast_nodes(code_ast.body, cell_name,
File "/usr/local/lib/python3.9/dist-packages/IPython/core/interactiveshell.py", line 3473, in run_ast_nodes
if (await self.run_code(code, result, async_=asy)):
File "/usr/local/lib/python3.9/dist-packages/IPython/core/interactiveshell.py", line 3553, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "<ipython-input-31-751d8af0bdfa>", line 134, in <cell line: 134>
model.fit([X_train_behavior, X_train_features], y_train, epochs=100, batch_size=1, verbose=1)
File "/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py", line 65, in error_handler
return fn(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 1685, in fit
tmp_logs = self.train_function(iterator)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 1284, in train_function
return step_function(self, iterator)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 1268, in step_function
outputs = model.distribute_strategy.run(run_step, args=(data,))
File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 1249, in run_step
outputs = model.train_step(data)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 1050, in train_step
y_pred = self(x, training=True)
File "/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py", line 65, in error_handler
return fn(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/training.py", line 558, in __call__
return super().__call__(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py", line 65, in error_handler
return fn(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/base_layer.py", line 1145, in __call__
outputs = call_fn(inputs, *args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py", line 96, in error_handler
return fn(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/functional.py", line 512, in call
return self._run_internal_graph(inputs, training=training, mask=mask)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/functional.py", line 669, in _run_internal_graph
outputs = node.layer(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py", line 65, in error_handler
return fn(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/engine/base_layer.py", line 1145, in __call__
outputs = call_fn(inputs, *args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/utils/traceback_utils.py", line 96, in error_handler
return fn(*args, **kwargs)
File "/usr/local/lib/python3.9/dist-packages/keras/layers/core/embedding.py", line 272, in call
out = tf.nn.embedding_lookup(self.embeddings, inputs)
Node: 'model_20/embedding_39/embedding_lookup_1'
indices[0] = -1 is not in [0, 27)
[[{{node model_20/embedding_39/embedding_lookup_1}}]] [Op:__inference_train_function_84956]