图来源：《如何选择机器学习的算法模型？》
用Python脚本梳理了一下图中的选择规则，就是用if...else...实现。
#!/usr/bin/env python
# -*- coding: utf-8 -*-
'''
n: 样本个数
predicting_a_category: True: 样本数据是类别预测，False: 数据不是类别预测
data_labeled: True: 数据有标签, False: 数据没有标签
# num_categories_known: True:类别数目是已知的; False: 类别数据不是已知的 # 对于不是类别预测的数据
# predicting_a_quality: True: 是量化数据预测; False: 不是量化数据预测
# few_feature_important: True: 少数的特征是重要的；False: 多数特征是重要的
# JustLooking: True: 只是看看数据；Fals: 不只是看看
# predicting_structure: True: 是结构化预测 False: 不是结构化预测(没有具体的模型建议)
# ModelWorking: True: 当前模型起作用，False: 当前模型不起来作用
'''
import argparse
# n = '10000'
# predicting_a_category = True
# data_labeled = True
# ModelWorking = True # 默认用True, 如果提供的模型，测试后不起作用，设置False
def read_args():
parser = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument('-n', '--num_samples', dest='num_samples',
metavar='INT', type=int, required=True,
help="set number of samples.")
parser.add_argument('--predicting_a_category', dest='predicting_a_category',
action='store_true', default=False,
help="set predict type is category or not, default: [False]")
parser.add_argument('--data_labeled', dest='data_labeled',
action='store_true', default=False,
help="set data is labeled or not, default: [False]")
parser.add_argument('--is_Text_Data', dest='is_Text_Data',
action='store_true', default=False,
help="set data format is text or not, default: [False]")
parser.add_argument('--num_categories_known', dest='num_categories_known',
action='store_true', default=False,
help="number of categories known or not, default: [False]")
parser.add_argument('--predicting_a_quality', dest='predicting_a_quality',
action='store_true', default=False,
help="set predict type is quality or not, default: [False]")
parser.add_argument('--few_feature_important', dest='few_feature_important',
action='store_true', default=False,
help="few feature should be inportant or not, default: [False]")
parser.add_argument('--JustLooking', dest='JustLooking',
action='store_true', default=False,
help="just looking or not, default: [False]")
parser.add_argument('--predicting_structure', dest='predicting_structure',
action='store_true', default=False,
help="set predict type is structure or not, default: [False]")
parser.add_argument('--ModelWorking', dest='ModelWorking',
action='store_true', default=False,
help="set default recommend model working or not, default: [False]")
args = parser.parse_args()
return args
def recommend_model(n, predicting_a_category, data_labeled,
is_Text_Data=False,
num_categories_known=True,
predicting_a_quality=False,
few_feature_important=False,
JustLooking=False,
predicting_structure=False,
ModelWorking=True):
model = ''
if n <= 50:
print('NOTE: need more samples! at least >=50')
exit(1) # 样本量太小，需要获取更多样本
else:
if predicting_a_category: # 如果是一个类别预测
if data_labeled: # 数据有标签
# classification，分类
if n < 100000: # n < 100,000
model = 'Linear SVC'
if not ModelWorking: # 如果该模型不起作用
if is_Text_Data: # 如果是文本数据
model = 'Naive Bayes'
else: # 如果不是文本数据
model = 'KNeighbors Classifier'
if not ModelWorking: # 如果该模型不起作用
model = 'SVC Ensemble Classifier'
else: # n >= 100,000
model = 'SGD Classifier'
if not ModelWorking:
model = 'kernel approximotion'
else:
# clustering, 聚类
if num_categories_known: # 如果类别数目已知
if n < 10000: # n < 10,000
model = 'KMeans'
if not ModelWorking:
model = 'Spectral Clustering GMM'
else:
model = 'MiniBatch KMeans'
else: # 如果类别数目未知
if n < 10000: # n < 10,000
model = 'MeanShift VBGMM'
else:
model = 'tough luck?' # 碰运气?
else:
if predicting_a_quality: # 如果是一个量化预测
if n < 100000: # n < 100,000
if few_feature_important: # 如果少数的特征是重要的
model = 'ElasticNet Lasso'
else:
model = "RidgeRegression SVR(kernel='liner')"
if not ModelWorking:
model = "SVR(kenel='rbf') EsembleRegression"
else: # n >= 100,000
model = 'SGD Regressor'
else: # 不是量化预测
if JustLooking: # 如果只是看以看？
# dimentionality reduction, 降维
model = 'Randomized PCA'
if not ModelWorking:
if n < 10000: # n < 10,000
model = 'Isomap Spectral Embedding'
if not ModelWorking:
model = 'LLE'
else: # n >= 10,000
model = 'kernel approximation'
else: # 不是只看看
if predicting_structure: # 如果是结构预测??
model = 'touch luck?' # 碰运气？？
if model == '':
print('Opps, NotFound any model for you!')
else:
print('Found model: "%s" for you~' % model)
return model
if __name__ == '__main__':
params = read_args()
recommend_model(params.num_samples, params.predicting_a_category, params.data_labeled,
is_Text_Data=params.is_Text_Data,
num_categories_known=params.num_categories_known,
predicting_a_quality=params.predicting_a_quality,
few_feature_important=params.few_feature_important,
JustLooking=params.JustLooking,
predicting_structure=params.predicting_structure,
ModelWorking=params.ModelWorking)
不同类别的算法：图片来源《如何学习机器学习算法》
简要介绍了有哪些"比较流行的机器学习算法"。
