python - 训练 LSTM 模型

Question

我正在尝试训练我的 lstm 模型，但在准确度、精度、召回率和 f1 分数上得到 0。我从kaggle下载了心脏病数据集。这是代码：

import tensorflow as tf
import pandas as pd
import numpy as np
from tensorflow.contrib import rnn
from sklearn.model_selection import train_test_split
from sklearn.metrics import f1_score, accuracy_score, recall_score, precision_score

heartt = pd.read_csv('heart.csv')

cols_to_norm = ['sex', 'cp', 'trestbps', 'chol', 'fbs', 'restecg', 'thalach', 'exang', 'oldpeak', 'slope', 'ca', 'thal']

heartt[cols_to_norm] = heartt[cols_to_norm].apply(lambda x: (x - x.min()) / (x.max() - x.min()))

sex_people = tf.feature_column.numeric_column('sex')
c_p = tf.feature_column.numeric_column('cp')
trest_bps = tf.feature_column.numeric_column('trestbps')
cholestrol = tf.feature_column.numeric_column('chol')
fb_s= tf.feature_column.numeric_column('fbs')
rest_ecg = tf.feature_column.numeric_column('restecg')
thala_ch = tf.feature_column.numeric_column('thalach')
ex_ang = tf.feature_column.numeric_column('exang')
old_peak = tf.feature_column.numeric_column('oldpeak')
slo_pe = tf.feature_column.numeric_column('slope')
c_a = tf.feature_column.numeric_column('ca')
tha_l = tf.feature_column.numeric_column('thal')
ag_e = tf.feature_column.numeric_column('age')

age_buckets = tf.feature_column.bucketized_column(ag_e, boundaries=[20,30,40,50,60,70,80])

feat_cols = [sex_people ,c_p, trest_bps ,cholestrol ,fb_s,rest_ecg,thala_ch ,ex_ang, old_peak, slo_pe,c_a, tha_l, age_buckets]

x_data = heartt.drop('target',axis=1)

x_data.info()

labels = heartt['target']

X_train,X_test,y_train,y_test = train_test_split(x_data, labels, test_size=0.2, shuffle=False, random_state=42)

epochs = 8
n_classes = 1
n_units = 200
n_features = 13
batch_size = 35

xplaceholder= tf.placeholder('float',[None,n_features])
yplaceholder = tf.placeholder('float')

def recurrent_neural_network_model():
    layer ={ 'weights': tf.Variable(tf.random_normal([n_units, n_classes])),'bias': tf.Variable(tf.random_normal([n_classes]))}

    x = tf.split(xplaceholder, n_features, 1)
    print(x)

    lstm_cell = rnn.BasicLSTMCell(n_units)

    outputs, states = rnn.static_rnn(lstm_cell, x, dtype=tf.float32)

    output = tf.matmul(outputs[-1], layer['weights']) + layer['bias']

    return output

def train_neural_network():
    logit = recurrent_neural_network_model()
    logit = tf.reshape(logit, [-1])

    cost = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=logit, labels=yplaceholder))
    optimizer = tf.train.AdamOptimizer().minimize(cost)

    with tf.Session() as sess:

        tf.global_variables_initializer().run()
        tf.local_variables_initializer().run()

        for epoch in range(epochs):
            epoch_loss = 0

            i = 0
            for i in range(int(len(X_train) / batch_size)):

                start = i
                end = i + batch_size

                batch_x = np.array(X_train[start:end])
                batch_y = np.array(y_train[start:end])

                _, c = sess.run([optimizer, cost], feed_dict={xplaceholder: batch_x, yplaceholder: batch_y})
                epoch_loss += c
                i += batch_size

            print('Epoch', epoch, 'completed out of', epochs, 'loss:', epoch_loss)

        pred = tf.round(tf.nn.sigmoid(logit)).eval({xplaceholder: np.array(X_test), yplaceholder: np.array(y_test)})
        f1 = f1_score(np.array(y_test), pred, average='macro')
        accuracy=accuracy_score(np.array(y_test), pred)
        recall = recall_score(y_true=np.array(y_test), y_pred= pred)
        precision = precision_score(y_true=np.array(y_test), y_pred=pred)
        print("F1 Score:", f1)
        print("Accuracy Score:",accuracy)
        print("Recall:", recall)
        print("Precision:", precision)


train_neural_network()

这是我得到的输出：

[<tf.Tensor 'split:0' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:1' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:2' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:3' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:4' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:5' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:6' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:7' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:8' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:9' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:10' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:11' shape=(?, 1) dtype=float32>, <tf.Tensor 'split:12' shape=(?, 1) dtype=float32>]
WARNING:tensorflow:From <ipython-input-15-5bc4f8465e4c>:8: BasicLSTMCell.__init__ (from tensorflow.python.ops.rnn_cell_impl) is deprecated and will be removed in a future version.
Instructions for updating:
This class is deprecated, please use tf.nn.rnn_cell.LSTMCell, which supports all the feature this cell currently has. Please replace the existing code with tf.nn.rnn_cell.LSTMCell(name='basic_lstm_cell').
Epoch 0 completed out of 8 loss: 1.00952459083328
Epoch 1 completed out of 8 loss: 3.3137323707244093e-06
Epoch 2 completed out of 8 loss: 1.6476146610239217e-09
Epoch 3 completed out of 8 loss: 2.08133817797794e-11
Epoch 4 completed out of 8 loss: 1.8306998712108724e-12
Epoch 5 completed out of 8 loss: 4.752560310897734e-13
Epoch 6 completed out of 8 loss: 2.238625324474169e-13
Epoch 7 completed out of 8 loss: 1.4679879558579696e-13
F1 Score: 0.0
Accuracy Score: 0.0
Recall: 0.0
Precision: 0.0
C:\ProgramData\Anaconda3\lib\site-packages\sklearn\metrics\classification.py:1135: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 in labels with no predicted samples.
  'precision', 'predicted', average, warn_for)
C:\ProgramData\Anaconda3\lib\site-packages\sklearn\metrics\classification.py:1137: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 in labels with no true samples.
  'recall', 'true', average, warn_for)
C:\ProgramData\Anaconda3\lib\site-packages\sklearn\metrics\classification.py:1137: UndefinedMetricWarning: Recall is ill-defined and being set to 0.0 due to no true samples.
  'recall', 'true', average, warn_for)

我对我在这里可能错的地方感到困惑。为什么我没有得到正确的准确率、精度、f1 分数和召回率？

标签： pythontensorflowmachine-learninglstmrecurrent-neural-network

Answer 1

我正在查看数据集，它看起来对 LSTM 模型来说不是问题。LSTM（以及一般的所有 RNN）旨在预测连续输出——它们是时间序列回归的神经网络等价物。我知道在某些情况下（例如使用 NLP 进行情感分析），您可以将 LSTM 应用于分类问题，但情况似乎并非如此。这些数据是“非时态的”，即数据集的每一行代表一个患者，数据的序列不携带任何信息。

当您需要具有“记忆”数据先前状态（例如时间序列）的模型时，使用 LSTM。如果您想应用 LSTM，我建议您更改数据集（您可以查看这个庞大的 ML 数据集列表）。否则，如果这是您感兴趣的数据集，则切换到前馈神经网络进行分类。为此，您可以查看我的个人 TensorFlow 教程，了解如何操作。