SVM模型 + 导出MLCore 最终方案

import pandas as pd
import numpy as np

# 读取 xlsx 文件
df = pd.read_excel('/Users/blibli/Documents/答辩数据/原始数据/T6发版本后修改/18939_20230116161751_处理中.xlsx')
df.fillna(0, inplace=True)

df = df.sample(n=8000)

# 连续数据标准化处理
df['VideoMusicIndex'] = (df['VideoMusicIndex'] - df['VideoMusicIndex'].mean()) / df['VideoMusicIndex'].std()
df['WordingCount_format100'] = (df['WordingCount_format100'] - df['WordingCount_format100'].mean()) / df[
    'WordingCount_format100'].std()
df['videoBitrate_format10000'] = (df['videoBitrate_format10000'] - df['videoBitrate_format10000'].mean()) / df[
    'videoBitrate_format10000'].std()
df['width_height_1080'] = (df['width_height_1080'] - df['width_height_1080'].mean()) / df[
    'width_height_1080'].std()

df_onehot = pd.get_dummies(df, columns=['EnterScene', 'VideoSource', 'LbsFlag', 'MusicType', 'professional_type',
                                        'sex_info', 'videoFrameRate_format', 'font_level'])

# 使用 pop 方法将 'postRet_format' 列从 DataFrame 中移除
postRet_format_column = df_onehot.pop('postRet_format')

# 使用 assign 方法将 'postRet_format' 列添加到 DataFrame 的末尾
df_onehot = df_onehot.assign(postRet_format=postRet_format_column)

df_onehot = df_onehot.head(13000) # 12000 精确度96%

from sklearn.model_selection import train_test_split

# print(df_onehot)

# print(df_onehot.columns)
print(df_onehot.shape[1])

# 将数据分成特征和目标
X = df_onehot.iloc[:, :(df_onehot.shape[1] - 1)]  # 填41时表示：选择前 41 列作为特征
y = df_onehot.iloc[:, (df_onehot.shape[1] - 1)]  # 选择第 42 作为目标

# print('y' + y)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.feature_selection import RFE
from sklearn.svm import SVC

# 将数据分成训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# 创建SVM模型
model = SVC(kernel='linear')

print('will begin')

# 使用RFE筛选出前10的特征量
# Recursive Feature Elimination(RFE)算法是数据竞赛中最为流行的特征筛选方案之一
selector = RFE(model, n_features_to_select=10)
X_new = selector.fit_transform(X, y)
print(X_new)

# 在训练数据上训练模型
model.fit(X_train, y_train)

print('will test')

# 在测试数据上测试模型
y_pred = model.predict(X_test)

# 计算模型的准确率
accuracy = accuracy_score(y_test, y_pred)
print('准确率：', accuracy)

print('params ！')
print(model.get_params())

print('hello')

import coremltools

print('begin mlmodel')

input_features =  np.array(X.columns)
print(input_features)
output_feature = "post"

input_model = coremltools.converters.sklearn.convert(model, input_features, output_feature)

# Set model metadata
input_model.author = 'binbinwang'
input_model.license = 'BSD'
input_model.short_description = 'Predict whether users will actually publish in the publishing interface.'

print('begin input_description')

# Set feature descriptions manually
# input_model.input_description['tv'] = 'Ads price spent on tv'
# input_model.input_description['radio'] = 'Ads price spent on radio'
# input_model.input_description['newspaper'] = 'Ads price spent on newspaper'

# Set the output descriptions
input_model.output_description['post'] = 'Users will trigger publishing behavior'

input_model.save("FinderPostModel.mlmodel")

print('Predict success')