Python利用神经网络解决非线性回归问题实例详解

本文实例讲述了Python利用神经网络解决非线性回归问题。分享给大家供大家参考,具体如下:

问题描述

现在我们通常使用神经网络进行分类,但是有时我们也会进行回归分析。
如本文的问题:
我们知道一个生物体内的原始有毒物质的量,然后对这个生物体进行治疗,向其体内注射一个物质,过一段时间后重新测量这个生物体内有毒物质量的多少。
因此,问题中有两个输入,都是标量数据,分别为有毒物质的量和注射物质的量,一个输出,也就是注射治疗物质后一段时间生物体的有毒物质的量。
数据如下图:

其中Dose of Mycotoxins 就是有毒物质,Dose of QCT就是治疗的药物。
其中蓝色底纹的数字就是输出结果。

一些说明

由于本文是进行回归分析,所以最后一层不进行激活,而直接输出。
本文程序使用sigmoid函数进行激活。
本文程序要求程序有一定的可重复性,隐含层可以指定。

另外,注意到
本文将使用数据预处理,也就是将数据减去均值再除以方差,否则使用sigmoid将会导致梯度消失。
因为数据比较大,比如200,这时输入200,当sigmoid函数的梯度就是接近于0了。
与此同时,我们在每一次激活前都进行BN处理,也就是batch normalize,中文可以翻译成规范化。
否则也会导致梯度消失的问题。与预处理情况相同。

程序

程序包括两部分,一部分是模型框架,一个是训练模型

第一部分:

# coding=utf-8
import numpy as np
def basic_forard(x, w, b):
  x = x.reshape(x.shape[0], -1)
  out = np.dot(x, w) + b
  cache = (x, w, b)
  return out, cache
def basic_backward(dout, cache):
  x, w, b = cache
  dout = np.array(dout)
  dx = np.dot(dout, w.T)
  # dx = np.reshape(dx, x.shape)
  # x = x.reshape(x.shape[0], -1)
  dw = np.dot(x.T, dout)
  db = np.reshape(np.sum(dout, axis=0), b.shape)
  return dx, dw, db
def batchnorm_forward(x, gamma, beta, bn_param):
  mode = bn_param['mode']
  eps = bn_param.get('eps', 1e-5)
  momentum = bn_param.get('momentum', 0.9)
  N, D = x.shape
  running_mean = bn_param.get('running_mean', np.zeros(D, dtype=x.dtype))
  running_var = bn_param.get('running_var', np.zeros(D, dtype=x.dtype))
  out, cache = None, None
  if mode == 'train':
    sample_mean = np.mean(x, axis=0)
    sample_var = np.var(x, axis=0)
    x_hat = (x - sample_mean) / (np.sqrt(sample_var + eps))
    out = gamma * x_hat + beta
    cache = (gamma, x, sample_mean, sample_var, eps, x_hat)
    running_mean = momentum * running_mean + (1 - momentum) * sample_mean
    running_var = momentum * running_var + (1 - momentum) * sample_var
  elif mode == 'test':
    scale = gamma / (np.sqrt(running_var + eps))
    out = x * scale + (beta - running_mean * scale)
  else:
    raise ValueError('Invalid forward batchnorm mode "%s"' % mode)
  bn_param['running_mean'] = running_mean
  bn_param['running_var'] = running_var
  return out, cache
def batchnorm_backward(dout, cache):
  gamma, x, u_b, sigma_squared_b, eps, x_hat = cache
  N = x.shape[0]
  dx_1 = gamma * dout
  dx_2_b = np.sum((x - u_b) * dx_1, axis=0)
  dx_2_a = ((sigma_squared_b + eps) ** -0.5) * dx_1
  dx_3_b = (-0.5) * ((sigma_squared_b + eps) ** -1.5) * dx_2_b
  dx_4_b = dx_3_b * 1
  dx_5_b = np.ones_like(x) / N * dx_4_b
  dx_6_b = 2 * (x - u_b) * dx_5_b
  dx_7_a = dx_6_b * 1 + dx_2_a * 1
  dx_7_b = dx_6_b * 1 + dx_2_a * 1
  dx_8_b = -1 * np.sum(dx_7_b, axis=0)
  dx_9_b = np.ones_like(x) / N * dx_8_b
  dx_10 = dx_9_b + dx_7_a
  dgamma = np.sum(x_hat * dout, axis=0)
  dbeta = np.sum(dout, axis=0)
  dx = dx_10
  return dx, dgamma, dbeta
# def relu_forward(x):
#   out = None
#   out = np.maximum(0,x)
#   cache = x
#   return out, cache
#
#
# def relu_backward(dout, cache):
#   dx, x = None, cache
#   dx = (x >= 0) * dout
#   return dx
def sigmoid_forward(x):
  x = x.reshape(x.shape[0], -1)
  out = 1 / (1 + np.exp(-1 * x))
  cache = out
  return out, cache
def sigmoid_backward(dout, cache):
  out = cache
  dx = out * (1 - out)
  dx *= dout
  return dx
def basic_sigmoid_forward(x, w, b):
  basic_out, basic_cache = basic_forard(x, w, b)
  sigmoid_out, sigmoid_cache = sigmoid_forward(basic_out)
  cache = (basic_cache, sigmoid_cache)
  return sigmoid_out, cache
# def basic_relu_forward(x, w, b):
#   basic_out, basic_cache = basic_forard(x, w, b)
#   relu_out, relu_cache = relu_forward(basic_out)
#   cache = (basic_cache, relu_cache)
#
#   return relu_out, cache
def basic_sigmoid_backward(dout, cache):
  basic_cache, sigmoid_cache = cache
  dx_sigmoid = sigmoid_backward(dout, sigmoid_cache)
  dx, dw, db = basic_backward(dx_sigmoid, basic_cache)
  return dx, dw, db
# def basic_relu_backward(dout, cache):
#   basic_cache, relu_cache = cache
#   dx_relu = relu_backward(dout, relu_cache)
#   dx, dw, db = basic_backward(dx_relu, basic_cache)
#
#   return dx, dw, db
def mean_square_error(x, y):
  x = np.ravel(x)
  loss = 0.5 * np.sum(np.square(y - x)) / x.shape[0]
  dx = (x - y).reshape(-1, 1)
  return loss, dx
class muliti_layer_net(object):
  def __init__(self, hidden_dim, input_dim=2, num_classes=2, weight_scale=0.01, dtype=np.float32, seed=None, reg=0.0, use_batchnorm=True):
    self.num_layers = 1 + len(hidden_dim)
    self.dtype = dtype
    self.reg = reg
    self.params = {}
    self.weight_scale = weight_scale
    self.use_batchnorm = use_batchnorm
    # init all parameters
    layers_dims = [input_dim] + hidden_dim + [num_classes]
    for i in range(self.num_layers):
      self.params['W' + str(i + 1)] = np.random.randn(layers_dims[i], layers_dims[i + 1]) * self.weight_scale
      self.params['b' + str(i + 1)] = np.zeros((1, layers_dims[i + 1]))
      if self.use_batchnorm and i < (self.num_layers - 1):
        self.params['gamma' + str(i + 1)] = np.ones((1, layers_dims[i + 1]))
        self.params['beta' + str(i + 1)] = np.zeros((1, layers_dims[i + 1]))
    self.bn_params = [] # list
    if self.use_batchnorm:
      self.bn_params = [{'mode': 'train'} for i in range(self.num_layers - 1)]
  def loss(self, X, y=None):
    X = X.astype(self.dtype)
    mode = 'test' if y is None else 'train'
    # compute the forward data and cache
    basic_sigmoid_cache = {}
    layer_out = {}
    layer_out[0] = X
    out_basic_forward, cache_basic_forward = {}, {}
    out_bn, cache_bn = {}, {}
    out_sigmoid_forward, cache_sigmoid_forward = {}, {}
    for lay in range(self.num_layers - 1):
      # print('lay: %f' % lay)
      W = self.params['W' + str(lay + 1)]
      b = self.params['b' + str(lay + 1)]
      if self.use_batchnorm:
        gamma, beta = self.params['gamma' + str(lay + 1)], self.params['beta' + str(lay + 1)]
        out_basic_forward[lay], cache_basic_forward[lay] = basic_forard(np.array(layer_out[lay]), W, b)
        out_bn[lay], cache_bn[lay] = batchnorm_forward(np.array(out_basic_forward[lay]), gamma, beta, self.bn_params[lay])
        layer_out[lay + 1], cache_sigmoid_forward[lay] = sigmoid_forward(np.array(out_bn[lay]))
         # = out_sigmoid_forward[lay]
      else:
        layer_out[lay+1], basic_sigmoid_cache[lay] = basic_sigmoid_forward(layer_out[lay], W, b)
    score, basic_cache = basic_forard(layer_out[self.num_layers-1], self.params['W' + str(self.num_layers)], self.params['b' + str(self.num_layers)])
    # print('Congratulations: Loss is computed successfully!')
    if mode == 'test':
      return score
    # compute the gradient
    grads = {}
    loss, dscore = mean_square_error(score, y)
    dx, dw, db = basic_backward(dscore, basic_cache)
    grads['W' + str(self.num_layers)] = dw + self.reg * self.params['W' + str(self.num_layers)]
    grads['b' + str(self.num_layers)] = db
    loss += 0.5 * self.reg * np.sum(self.params['W' + str(self.num_layers)] * self.params['b' + str(self.num_layers)])
    dbn, dsigmoid = {}, {}
    for index in range(self.num_layers - 1):
      lay = self.num_layers - 1 - index - 1
      loss += 0.5 * self.reg * np.sum(self.params['W' + str(lay + 1)] * self.params['b' + str(lay + 1)])
      if self.use_batchnorm:
        dsigmoid[lay] = sigmoid_backward(dx, cache_sigmoid_forward[lay])
        dbn[lay], grads['gamma' + str(lay + 1)], grads['beta' + str(lay + 1)] = batchnorm_backward(dsigmoid[lay], cache_bn[lay])
        dx, grads['W' + str(lay + 1)], grads['b' + str(lay + 1)] = basic_backward(dbn[lay], cache_basic_forward[lay])
      else:
        dx, dw, db = basic_sigmoid_backward(dx, basic_sigmoid_cache[lay])
    for lay in range(self.num_layers):
      grads['W' + str(lay + 1)] += self.reg * self.params['W' + str(lay + 1)]
    return loss, grads
def sgd_momentum(w, dw, config=None):
  if config is None: config = {}
  config.setdefault('learning_rate', 1e-2)
  config.setdefault('momentum', 0.9)
  v = config.get('velocity', np.zeros_like(w))
  v = config['momentum'] * v - config['learning_rate'] * dw
  next_w = w + v
  config['velocity'] = v
  return next_w, config
class Solver(object):
  def __init__(self, model, data, **kwargs):
    self.model = model
    self.X_train = data['X_train']
    self.y_train = data['y_train']
    self.X_val = data['X_val']
    self.y_val = data['y_val']
    self.update_rule = kwargs.pop('update_rule', 'sgd_momentum')
    self.optim_config = kwargs.pop('optim_config', {})
    self.lr_decay = kwargs.pop('lr_decay', 1.0)
    self.batch_size = kwargs.pop('batch_size', 100)
    self.num_epochs = kwargs.pop('num_epochs', 10)
    self.weight_scale = kwargs.pop('weight_scale', 0.01)
    self.print_every = kwargs.pop('print_every', 10)
    self.verbose = kwargs.pop('verbose', True)
    if len(kwargs) > 0:
      extra = ', '.join('"%s"' % k for k in kwargs.keys())
      raise ValueError('Unrecognized argements %s' % extra)
    self._reset()
  def _reset(self):
    self.epoch = 100
    self.best_val_acc = 0
    self.best_params = {}
    self.loss_history = []
    self.train_acc_history = []
    self.val_acc_history = []
    self.optim_configs = {}
    for p in self.model.params:
      d = {k: v for k, v in self.optim_config.items()}
      self.optim_configs[p] = d
  def _step(self):
    loss, grads = self.model.loss(self.X_train, self.y_train)
    self.loss_history.append(loss)
    for p, w in self.model.params.items():
      dw = grads[p]
      config = self.optim_configs[p]
      next_w, next_config = sgd_momentum(w, dw, config)
      self.model.params[p] = next_w
      self.optim_configs[p] = next_config
    return loss
  def train(self):
    min_loss = 100000000
    num_train = self.X_train.shape[0]
    iterations_per_epoch = max(num_train / self.batch_size, 1)
    num_iterations = self.num_epochs * iterations_per_epoch
    for t in range(int(num_iterations)):
      loss = self._step()
      if self.verbose:
#         print(self.loss_history[-1])
        pass
      if loss < min_loss:
        min_loss = loss
        for k, v in self.model.params.items():
          self.best_params[k] = v.copy()
    self.model.params = self.best_params

第二部分

import numpy as np
# import data
dose_QCT = np.array([0, 5, 10, 20])
mean_QCT, std_QCT = np.mean(dose_QCT), np.std(dose_QCT)
dose_QCT = (dose_QCT - mean_QCT ) / std_QCT
dose_toxins = np.array([0, 0.78125, 1.5625, 3.125, 6.25, 12.5, 25, 50, 100, 200])
mean_toxins, std_toxins = np.mean(dose_toxins), np.std(dose_toxins)
dose_toxins = (dose_toxins - mean_toxins ) / std_toxins
result = np.array([[0, 4.037, 7.148, 12.442, 18.547, 25.711, 34.773, 62.960, 73.363, 77.878],
          [0, 2.552, 4.725, 8.745, 14.436, 21.066, 29.509, 55.722, 65.976, 72.426],
          [0, 1.207, 2.252, 4.037, 7.148, 11.442, 17.136, 34.121, 48.016, 60.865],
          [0, 0.663, 1.207, 2.157, 3.601, 5.615, 8.251, 19.558, 33.847, 45.154]])
mean_result, std_result = np.mean(result), np.std(result)
result = (result - mean_result ) / std_result
# create the train data
train_x, train_y = [], []
for i,qct in enumerate(dose_QCT):
  for j,toxin in enumerate(dose_toxins):
    x = [qct, toxin]
    y = result[i, j]
    train_x.append(x)
    train_y.append(y)
train_x = np.array(train_x)
train_y = np.array(train_y)
print(train_x.shape)
print(train_y.shape)
import layers_regression
small_data = {'X_train': train_x,
       'y_train': train_y,
       'X_val': train_x,
       'y_val': train_y,}
batch_size = train_x.shape[0]
learning_rate = 0.002
reg = 0
model = layers_regression.muliti_layer_net(hidden_dim=[5,5], input_dim=2, num_classes=1, reg=reg, dtype=np.float64)
solver = layers_regression.Solver(model, small_data, print_every=0, num_epochs=50000, batch_size=batch_size, weight_scale=1,
                 update_rule='sgd_momentum', optim_config={'learning_rate': learning_rate})
print('Please wait several minutes!')
solver.train()
# print(model.params)
best_model = model
print('Train process is finised')
import matplotlib.pyplot as plt
# %matplotlib inline
plt.plot(solver.loss_history, '.')
plt.title('Training loss history')
plt.xlabel('Iteration')
plt.ylabel('Training loss')
plt.show()
# predict the training_data
predict = best_model.loss(train_x)
predict = np.round(predict * std_result + mean_result, 1)
print('Predict is ')
print('{}'.format(predict.reshape(4, 10)))
# print('{}'.format(predict))
# observe the error between the predict after training with ground truth
result = np.array([[0, 4.037, 7.148, 12.442, 18.547, 25.711, 34.773, 62.960, 73.363, 77.878],
          [0, 2.552, 4.725, 8.745, 14.436, 21.066, 29.509, 55.722, 65.976, 72.426],
          [0, 1.207, 2.252, 4.037, 7.148, 11.442, 17.136, 34.121, 48.016, 60.865],
          [0, 0.663, 1.207, 2.157, 3.601, 5.615, 8.251, 19.558, 33.847, 45.154]])
result = result.reshape(4, 10)
predict = predict.reshape(4, 10)
error = np.round(result - predict, 2)
print('error between predict and real data')
print(error)
print('The absulate error in all data is %f' % np.sum(np.abs(error)))
print('The mean error in all data is %f' % np.mean(np.abs(error)))
# figure the predict map in 3D
x_1 = (np.arange(0, 20, 0.1) - mean_QCT) / std_QCT
x_2 = (np.arange(0, 200, 1) - mean_toxins) / std_toxins
x_test = np.zeros((len(x_1)*len(x_2), 2))
index = 0
for i in range(len(x_1)):
  for j in range(len(x_2)):
    x_test[int(index), 0] = x_1[int(i)]
    x_test[int(index), 1] = x_2[int(j)]
    index += 1
test_pred = best_model.loss(x_test)
predict = np.round(test_pred * std_result + mean_result, 3)
from mpl_toolkits.mplot3d import Axes3D
x_1, x_2 = np.meshgrid(x_1 * std_QCT + mean_QCT, x_2 * std_toxins + mean_toxins)
figure = plt.figure()
ax = Axes3D(figure)
predict = predict.reshape(len(x_1), len(x_2))
ax.plot_surface(x_1, x_2, predict, rstride=1, cstride=1, cmap='rainbow')
plt.show()
# 最后本文将进行一些预测,但预测效果不是很好
# question 2: predict with given
dose_QCT_predict = np.ravel(np.array([7.5, 15]))
dose_QCT_predict_ = (dose_QCT_predict - mean_QCT)/ std_QCT
dose_toxins_predict = np.array([0, 0.78125, 1.5625, 3.125, 6.25, 12.5, 25, 50, 100, 200])
dose_toxins_predict_ = (dose_toxins_predict - mean_toxins) / std_toxins
test = []
for i,qct in enumerate(dose_QCT_predict):
  for j,toxin in enumerate(dose_toxins_predict):
    x = [qct, toxin]
    test.append(x)
test = np.array(test)
print('Please look at the test data:')
print(test)
test = []
for i,qct in enumerate(dose_QCT_predict_):
  for j,toxin in enumerate(dose_toxins_predict_):
    x = [qct, toxin]
    test.append(x)
test = np.array(test)
test_pred = best_model.loss(test)
predict = np.round(test_pred * std_result + mean_result, 1)
print(predict.reshape(2, 10))

更多关于Python相关内容感兴趣的读者可查看本站专题:《Python数学运算技巧总结》、《Python数据结构与算法教程》、《Python函数使用技巧总结》、《Python字符串操作技巧汇总》及《Python入门与进阶经典教程》

希望本文所述对大家Python程序设计有所帮助。

(0)

相关推荐

  • Python解决pip install时出现的Could not fetch URL问题

    前言 使用python直接使用pip install xx时,出现 Could not fetch URL https://pypi.python.org/simple/requests/: There was a problem confirming the ssl certificate: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify faile 这种错误,可以用以下方法解决.本人也是拜访无数帖子,尝试了几个小时弄好的,希望可以给大家

  • 完美解决Python 2.7不能正常使用pip install的问题

    主要原因是用户目录编码使用了中文,解决方法如下: pip安装python包会加载我的用户目录,我的用户目录恰好是中文的,ascii不能编码. 解决办法是: python目录 Python27\Lib\site-packages 建一个文件sitecustomize.py 内容写: import sys sys.setdefaultencoding('gb2312') python会自动运行这个文件. 以上这篇完美解决Python 2.7不能正常使用pip install的问题就是小编分享给大家的

  • 解决python xlrd无法读取excel文件的问题

    读取文件时报错: xlrd.biffh.XLRDError: Unsupported format, or corrupt file: Expected BOF record; found '<?xml' 该xls文件在打开时确实会警告该文件与扩展名格式不一致.用文本编辑器打开该xls文件查看,发现确实不是xls文件,而是xml文件被保存为了xls文件. 解决办法:将文件后缀名改为.xml,作为xml文件读入. 改名代码: import os os.rename(u"D:\PycharmP

  • 解决Python 命令行执行脚本时,提示导入的包找不到的问题

    在Pydev能正常执行的脚本,在导出后在命令行执行,通常会报自己写的包导入时找不到. 一:报错原因 在PyDev中,test.py 中导入TestUserCase里面的py文件时,会写: from TestUserCase.Test import Test 这在命Pydev中没错,但是在命令行中确出现错误. 因为在PyDev中默认路径除了当前文件所在路劲外,还有工程根路径.而在命令行中只有当前文件所在路径.TestUserCase是test.py文件的上一级目录,所以在命令行中test.py i

  • 解决Python中pandas读取*.csv文件出现编码问题

    1.问题 在使用Python中pandas读取csv文件时,由于文件编码格式出现以下问题: Traceback (most recent call last): File "pandas\_libs\parsers.pyx", line 1134, in pandas._libs.parsers.TextReader._convert_tokens File "pandas\_libs\parsers.pyx", line 1240, in pandas._libs

  • Python利用神经网络解决非线性回归问题实例详解

    本文实例讲述了Python利用神经网络解决非线性回归问题.分享给大家供大家参考,具体如下: 问题描述 现在我们通常使用神经网络进行分类,但是有时我们也会进行回归分析. 如本文的问题: 我们知道一个生物体内的原始有毒物质的量,然后对这个生物体进行治疗,向其体内注射一个物质,过一段时间后重新测量这个生物体内有毒物质量的多少. 因此,问题中有两个输入,都是标量数据,分别为有毒物质的量和注射物质的量,一个输出,也就是注射治疗物质后一段时间生物体的有毒物质的量. 数据如下图: 其中Dose of Myco

  • python生成二维码的实例详解

    python生成二维码的实例详解 版本相关 操作系统:Mac OS X EI Caption Python版本:2.7 IDE:Sublime Text 3 依赖库 Python生成二维码需要的依赖库为PIL和QRcode. 坑爹的是,百度了好久都没有找到PIL,不知道是什么时候改名了,还是其他原因,pillow就是传说中的PIL. 安装命令:sudo pip install pillow.sudo pip install qrcode 验证是否安装成功,使用命令from PIL import

  • Python生成短uuid的方法实例详解

    python的uuid都是32位的,比较长,处理起来效率比较低, 本算法利用62个可打印字符,通过随机生成32位UUID,由于UUID都为十六进制,所以将UUID分成8组,每4个为一组,然后通过模62操作,结果作为索引取出字符, 最后生成的Uuid,只有8位,代码如下: uuid4,可以换成uuid1 from uuid import uuid4 uuidChars = ("a", "b", "c", "d", "e

  • python 利用pyttsx3文字转语音过程详解

    这篇文章主要介绍了python 利用pyttsx3文字转语音过程详解,文中通过示例代码介绍的非常详细,对大家的学习或者工作具有一定的参考学习价值,需要的朋友可以参考下 # -*- coding: utf-8 -*- import pyttsx3 engine = pyttsx3.init() with open("all.txt",'r',encoding='utf-8') as f: while 1: line = f.readline() print(line, end = '')

  • python中的二维列表实例详解

    1. 使用输入值初始化列表 nums = [] rows = eval(input("请输入行数:")) columns = eval(input("请输入列数:")) for row in range(rows): nums.append([]) for column in range(columns): num = eval(input("请输入数字:")) nums[row].append(num) print(nums) 输出结果为: 请

  • python实现人性化显示金额数字实例详解

    我们在开发过程中,有时候需要把float,int型等数字作为金额类型数字显示会出现很多问题,比如float会显示成 965868.4599999,int型没有小数位等各种各样的问题.我们需要进行转换显示,才能保证阅读人性化. 方法一: 这里只贴上主要代码: # 金额人性化 def humanized_amount(self, *args, **kwargs):     """     金额人性化,保留二位小数,再进行人性化显示     compel强制二位,默认True,  

  • python爬取天气数据的实例详解

    就在前几天还是二十多度的舒适温度,今天一下子就变成了个位数,小编已经感受到冬天寒风的无情了.之前对获取天气都是数据上的搜集,做成了一个数据表后,对温度变化的感知并不直观.那么,我们能不能用python中的方法做一个天气数据分析的图形,帮助我们更直接的看出天气变化呢? 使用pygal绘图,使用该模块前需先安装pip install pygal,然后导入import pygal bar = pygal.Line() # 创建折线图 bar.add('最低气温', lows) #添加两线的数据序列 b

  • python压包的概念及实例详解

    对于一些分解后的元素,我们也是有重新归类的需要.那么我们把解包的恢复过程,叫做压包.这里要用到zip函数的方法,对元素重新进行打包处理,在之前的学习中我们已经对zip函数有所接触.下面我们就python压包的概念.方法进行介绍,然后带来相关的实例使用. 1.概念 压包是解包的逆过程,用zip函数实现. 2.方法 (1)zip() 函数用于将可迭代的对象作为参数,将对象中对应的元素打包成一个个元组,然后返回由这些元组组成的对象(Python3). (2)如果各个迭代器的元素个数不一致,则返回列表长

  • Python机器学习应用之决策树分类实例详解

    目录 一.数据集 二.实现过程 1 数据特征分析 2 利用决策树模型在二分类上进行训练和预测 3 利用决策树模型在多分类(三分类)上进行训练与预测 三.KEYS 1 构建过程 2 划分选择 3 重要参数 一.数据集 小企鹅数据集,提取码:1234 该数据集一共包含8个变量,其中7个特征变量,1个目标分类变量.共有150个样本,目标变量为 企鹅的类别 其都属于企鹅类的三个亚属,分别是(Adélie, Chinstrap and Gentoo).包含的三种种企鹅的七个特征,分别是所在岛屿,嘴巴长度,

  • python open函数中newline参数实例详解

    目录 问题的由来 具体实例 总结 问题的由来 我在读pythoncsv模块文档 看到了这样一句话 如果 csvfile 是文件对象,则打开它时应使用 newline=‘’.其备注:如果没有指定 newline=‘’,则嵌入引号中的换行符将无法正确解析,并且在写入时,使用 \r\n 换行的平台会有多余的 \r 写入.由于 csv 模块会执行自己的(通用)换行符处理,因此指定 newline=‘’ 应该总是安全的. 我就在思考open函数中的newline参数的作用,因为自己之前在使用open函数时

随机推荐