构建seq2seq模型训练已脱敏数据集，Encoder/Decoder的选择LSTM，评估指标：bleu-4 (from nltk.translate.bleu_score import corpus_bleu)。数据集类似以下格式【index,description,diagnosis 0,14 108 28 30 15 13 294 29 20 18 23 21 25 32 16 14 39 27 14 47 46 69 70 11 24 42 26 37 61 24 10 79 46 62 19 13 31 95 19 28 20 18 10 22 12 38 41 17 23 21 36 53 25 10,22 12 38 41 17 81 10】，写出其python代码

好的，下面是我给您提供的代码，其中包括数据处理、模型定义、训练和评估等部分：

import numpy as np
import pandas as pd
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from keras.models import Model
from keras.layers import Input, LSTM, Dense, Embedding
from keras.callbacks import ModelCheckpoint
from nltk.translate.bleu_score import corpus_bleu

# 数据加载
data = pd.read_csv("data.csv", encoding="utf-8")

# 将数据划分为训练集和验证集
train_size = int(len(data) * 0.8)
train_data = data[:train_size]
val_data = data[train_size:]

# 训练集和验证集的输入和输出
train_input = train_data['description'].values
train_output = train_data['diagnosis'].values
val_input = val_data['description'].values
val_output = val_data['diagnosis'].values

# 创建标记器
tokenizer = Tokenizer(oov_token='<UNK>')
tokenizer.fit_on_texts(np.concatenate([train_input, train_output]))

# 训练集和验证集的输入和输出序列
train_input_seq = tokenizer.texts_to_sequences(train_input)
train_output_seq = tokenizer.texts_to_sequences(train_output)
val_input_seq = tokenizer.texts_to_sequences(val_input)
val_output_seq = tokenizer.texts_to_sequences(val_output)

# 获取最大序列长度
max_input_len = max(len(seq) for seq in train_input_seq)
max_output_len = max(len(seq) for seq in train_output_seq)

# 填充序列
train_input_seq = pad_sequences(train_input_seq, maxlen=max_input_len, padding='post')
train_output_seq = pad_sequences(train_output_seq, maxlen=max_output_len, padding='post')
val_input_seq = pad_sequences(val_input_seq, maxlen=max_input_len, padding='post')
val_output_seq = pad_sequences(val_output_seq, maxlen=max_output_len, padding='post')

# 定义模型
latent_dim = 256

encoder_inputs = Input(shape=(max_input_len,))
encoder_embedding = Embedding(len(tokenizer.word_index) + 1, latent_dim, mask_zero=True)(encoder_inputs)
encoder_lstm = LSTM(latent_dim, return_state=True)
encoder_outputs, state_h, state_c = encoder_lstm(encoder_embedding)
encoder_states = [state_h, state_c]

decoder_inputs = Input(shape=(max_output_len,))
decoder_embedding = Embedding(len(tokenizer.word_index) + 1, latent_dim, mask_zero=True)(decoder_inputs)
decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True)
decoder_outputs, _, _ = decoder_lstm(decoder_embedding, initial_state=encoder_states)
decoder_dense = Dense(len(tokenizer.word_index) + 1, activation='softmax')
decoder_outputs = decoder_dense(decoder_outputs)

model = Model([encoder_inputs, decoder_inputs], decoder_outputs)

# 编译模型
model.compile(optimizer='rmsprop', loss='sparse_categorical_crossentropy')

# 定义回调函数
checkpoint = ModelCheckpoint('model.h5', save_best_only=True)

# 训练模型
batch_size = 64
epochs = 50
history = model.fit([train_input_seq, train_output_seq[:, :-1]], train_output_seq[:, 1:],
                    batch_size=batch_size,
                    epochs=epochs,
                    validation_data=([val_input_seq, val_output_seq[:, :-1]], val_output_seq[:, 1:]),
                    callbacks=[checkpoint])

# 评估模型
test_input = val_input_seq[:10]
test_output = val_output_seq[:10]

pred_output = []
for i in range(len(test_input)):
    input_seq = test_input[i:i+1]
    decoded_seq = np.zeros((1, max_output_len))
    decoded_seq[0, 0] = tokenizer.word_index['<s>']
    for j in range(1, max_output_len):
        output = model.predict([input_seq, decoded_seq]).argmax(axis=-1)
        decoded_seq[0, j] = output[0, j-1]
        if output[0, j-1] == tokenizer.word_index['</s>']:
            break
    pred_output.append(decoded_seq[0])

pred_output = np.array(pred_output)
test_output = test_output[:, 1:]
pred_output = pred_output[:, :-1]

print('BLEU-4 score:', corpus_bleu(test_output, pred_output))

这段代码是一个完整的Seq2Seq模型训练和评估的示例，其中包括数据处理、模型定义、训练和评估等部分。您可以根据自己的具体情况对代码进行修改和调整。