0

我对 keras 很陌生，并试图构建一个模型，该模型将列表作为输入并返回 1 到 16 之间的数字（或 0 到 15，我有 16 个类）

到目前为止，这是我的代码：

import numpy as np
from tensorflow.keras.layers.experimental.preprocessing import TextVectorization
from tensorflow.keras import layers
from tensorflow import keras
import tensorflow as tf

vectorizer = TextVectorization(output_mode = "int")

'''
vocab_data = np.array(["In animus fert nova"])

vectorizer.adapt(vocab_data) #wichtig

print(vectorizer(vocab_data))
'''
# 2 Möglichkeiten für die ersten 4 Metren --> 2^4 = 16 verschiedene Hexameter
num_Hexameter = 16

inputs = keras.Input(shape = (None,1), dtype=tf.int64)
x = layers.Dense(30)(inputs)
x = layers.Dense(20)(x)
x = layers.Dense(15)(x)
x = layers.Dense(10)(x)
x = layers.Dense(5)(x)
outputs = layers.Dense(num_Hexameter, activation = "softmax")(x)

model = keras.Model(inputs = inputs, outputs = outputs)

model.summary()

model.compile(optimizer = keras.optimizers.RMSprop(learning_rate=1e-3),
    loss=keras.losses.SparseCategoricalCrossentropy(), metrics=["accuracy"])

test_train_data_raw = np.array([["In nova fert animus mutatas dicere formas"], 
["corpora di coeptis nam vos mutastis et illas"], 
["adspirate meis primaque ab origine mundi"],
["ad mea perpetuum deducite tempora carmen"]])
vectorizer.adapt(test_train_data_raw)
test_train_data = vectorizer(test_train_data_raw)
test_train_data_lbl = np.array([[0, 0, 0, 0, 0, 0, 0, 3],[0, 0, 0, 0, 0, 0, 0, 7], [0, 0, 0, 0, 0, 0, 0, 10], [0, 0, 0, 0, 0, 0, 0, 2]])
test_train_data_lbl = np.array([3, 7, 10, 2])
print(test_train_data_lbl.shape)
history = model.fit(test_train_data, test_train_data_lbl, batch_size = 2, epochs = 10)

print(history.history)
test = vectorizer(np.array([["In nova fert animus mutatas dicere formas"]]))
print(test)
print(model.predict(test).shape)

问题是如果我不test_train_data_lbl = np.array([3, 7, 10, 2])注释该行，它会给我错误：ValueError: Shape mismatch: The shape of labels (received (2, 1)) should equal the shape of logits except for the last dimension (received (2, 8, 16))

注释该行不会导致错误，但结果model.predict(test)将是一个带有 shape 的数组(1, 7, 16)。我知道 7 来自我拥有的 7 个单词和 16 来自类的数量，但我需要这个形状(1, 16)，以便我可以预测该行将在哪个类中。

我现在也必须减少训练数据，但我首先想让模型在生成训练数据之前没有错误地工作。

4

0 回答 0