基于机器学习的音乐风格分类

1 任务介绍

基于机器学习的音乐风格分类，包含数据的导入，特征选择，机器学习模型，超参数调整。
数据集连接：https://download.csdn.net/download/ww596520206/87860225

2 数据导入

data文件夹包含了所有数据，其中文件名为类别名称，可以发现共有6类，分别为：classical、country、edm、jazz、rap、rock，读取文件夹，计算每个类别的数目如下所示

import pandas as pd
import numpy as np
from sklearn.utils import shuffle
from sklearn.preprocessing import MinMaxScaler
from sklearn.svm import SVC
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier, AdaBoostClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import accuracy_score, recall_score, f1_score, confusion_matrix
import matplotlib.pyplot as plt

classical = pd.read_csv("data/classical_csv.csv")
country = pd.read_csv("data/country_csv.csv")
edm = pd.read_csv("data/edm_csv.csv")
jazz = pd.read_csv("data/jazz_csv.csv")
rap = pd.read_csv("data/rap_csv.csv")
rock = pd.read_csv("data/rock_csv.csv")

print("classical num is "+str(len(classical)))
print("country num is "+str(len(country)))
print("edm num is "+str(len(edm)))
print("jazz num is "+str(len(jazz)))
print("rap num is "+str(len(rap)))
print("rock num is "+str(len(rock)))

classical num is 2119
country num is 2011
edm num is 2043
jazz num is 2178
rap num is 2162
rock num is 2035

2 数据处理

2.1 数据预处理，为每个数据添加类别

classical.insert(loc=10, column='class', value=0)
country.insert(loc=10, column='class', value=1)
edm.insert(loc=10, column='class', value=2)
jazz.insert(loc=10, column='class', value=3)
rap.insert(loc=10, column='class', value=4)
rock.insert(loc=10, column='class', value=5)
all_data = pd.concat([classical, country, edm, jazz, rap, rock])

2.2 特征： Unnamed: 0是每个歌曲的编号，对分类没有作用，可以丢掉

all_data.drop(["Unnamed: 0"],axis=1, inplace=True)

2.3 计算每个特征的方差，剔除掉方差过小的特征

all_data.std(axis=0)
all_data.drop(["speechiness"],axis=1, inplace=True)

最后保留的特征为：acousticness、danceability 、energy 、instrumentalness、key、loudness、tempo、valence ，共计8个。

3 特征分析

3.1 绘制每个特征的概率密度函数

for i, col in enumerate(all_data.columns[:-1]):
    plt.subplot(2, 4, i + 1)
    one_data = all_data.loc[:, col]
    one_data.plot(kind="kde")
    plt.title(col)
plt.show()

3.2 绘制每个特征的直方图

for i, col in enumerate(all_data.columns[:-1]):
    plt.subplot(2, 4, i + 1)
    feature_data = all_data.loc[:, col]
    feature_data.plot(kind="hist")
    plt.title(col)
plt.show()

4 模型训练和验证

4.1 对于训练集划分训练集和验证集，训练集占70%，验证集占30%，并且前11列为特征列，最后一列为标签列

all_data = shuffle(all_data)
train_num = int(len(all_data) * 0.7)
train_features, train_labels = all_data.iloc[:train_num, :-1].values, all_data.iloc[:train_num, -1].values
val_features, val_labels = all_data.iloc[train_num:, :-1].values, all_data.iloc[train_num:, -1].values

4.2 数据归一化

scaler = MinMaxScaler()
scaler.fit(train_features)
train_features = scaler.transform(train_features)
val_features = scaler.transform(val_features)

4.3 搭建模型，使用随机森林模型,使用cpu计算，随机森林是集成模型，包含了好多树模型。

model = RandomForestClassifier()
model.fit(train_features, train_labels) # 训练模型
train_pred = model.predict(train_features) # 在训练集测试
val_pred = model.predict(val_features) # 在验证集测试
print("train acc: {}".format(accuracy_score(train_labels, train_pred)))
print("val acc:{}".format(accuracy_score(val_labels, val_pred)))

4.4 使用交叉验证评估参数

score_list = []
cv_num = 10
for n_estimators in [25, 50, 70]:
    model = RandomForestClassifier(n_estimators=n_estimators)
    score = cross_val_score(model, train_features, train_labels, cv=cv_num)
    score_list.append(score)

plt.plot(range(cv_num), score_list[0],
         range(cv_num), score_list[1],
         range(cv_num), score_list[2],)
plt.legend(["n=25","n=50","n=70"])
plt.show()