Question

我有一个如下所示的数据集：

class Release(db.Model):
    __tablename__ = 'releases'
    id = db.Column(db.Integer, primary_key=True)
    platform_id=db.Column(db.Integer, db.ForeignKey('platforms.id'))
    name = db.Column(db.String(20), unique=True)
    builds = db.relationship('ReleaseBuilds', cascade='all,delete', lazy='dynamic', order_by="desc(ReleaseBuilds.date_created)")

class ReleaseBuilds(db.Model):
    __tablename__='release_builds'
    id = db.Column(db.Integer, primary_key=True)
    release_id = db.Column(db.Integer, db.ForeignKey('releases.id'))
    name = db.Column(db.String(150), nullable=False)
    artifacts = db.relationship('ReleaseBuildArtifacts', cascade='all,delete', backref='builds', lazy='dynamic')
    deployments = db.relationship('Deployments', cascade='all,delete', lazy='dynamic')
    tests = db.relationship('Test', cascade='delete', lazy='dynamic')

class ReleaseBuildArtifacts(db.Model):
    __tablename__='release_build_artifacts'
    id = db.Column(db.Integer, primary_key=True)
    release_build_id = db.Column(db.Integer, db.ForeignKey('release_builds.id'))
    application_id = db.Column(db.Integer, db.ForeignKey('applications.id'))
    rpm = db.Column(db.String(300))
    build = db.relationship('ReleaseBuilds')
    application = db.relationship('Application')

class Deployments(db.Model):
    __tablename__ = 'deployments'
    release_build_id = db.Column(db.Integer, db.ForeignKey('release_builds.id'), primary_key=True)
    environment_id = db.Column(db.Integer, db.ForeignKey('environments.id'), primary_key=True)
    date_deployed = db.Column(db.DateTime(timezone=False), default=datetime.datetime.utcnow)
    environment = db.relationship('Environment', foreign_keys=[environment_id])

class TestType(db.Model):
    __tablename__ = 'test_types'
    id = db.Column(db.Integer, primary_key=True)
    name = db.Column(db.String(50), unique=True)

class Test(db.Model):
    __tablename__ = 'tests'
    id = db.Column(db.Integer, primary_key=True)
    release_build_id = db.Column(db.Integer, db.ForeignKey('release_builds.id'), nullable=False)
    environment_id = db.Column(db.Integer, db.ForeignKey('environments.id'), nullable=False)
    test_type_id = db.Column(db.Integer, db.ForeignKey('test_types.id'))
    name = db.Column(db.String(300))
    environments = db.relationship('Environment', foreign_keys=[environment_id])
    results = db.relationship('TestResult', cascade='all,delete', lazy='dynamic')

    __table_args__ = (
        ForeignKeyConstraint(['release_build_id', 'environment_id'],['deployments.release_build_id', 'deployments.environment_id']),
        )

class TestResult(db.Model):
    __tablename__ = 'test_results'
    id = db.Column(db.Integer, primary_key=True)
    test_id = db.Column(db.Integer, db.ForeignKey('tests.id'), nullable=False)
    name = db.Column(db.String(500))
    passed = db.Column(db.Boolean)

可能有双重日期，日期之间可能存在差距（缺少日期）

每天我想计算过去30天内发生的行数（个案数）。

我成功地通过以下方式做到了这一点：

date, brandname, status, case number
2017-01-01, x1, closed, 12345
2017-01-01, x2, closed, 12345
2017-01-01, x3, closed, 12345
2017-01-02, x4, open, 7864
2017-01-03, x5, open, 78642
...

这给了我所需的确切结果：

dataframe <- 
  structure(list(date = structure(c(17167, 17167, 17167, 17168, 
  17169), class = "Date"), brandname = c("x1", "x2", "x3", "x4", 
  "x5"), status = c("closed", "closed", "closed", "open", "open"
  ), `case number` = c(12345L, 12345L, 12345L, 7864L, 78642L)), .Names = c("date", 
  "brandname", "status", "case number"), row.names = c(NA, -5L), class = "data.frame")

for(i in 1:nrow(dataframe)) {
   frame <- subset(dataframe, date > dataframe$date[i] - 30L & 
                              date < dataframe$date[i])
   dataframe$count[i] <- nrow(frame)
}

但我相信有更有效的方法可以做到这一点。有人可以帮忙吗？

Answer 1

这可以使用 non-equi join 来解决：

library(data.table)
DT <- as.data.table(dataframe)
DT[, count := DT[DT[, .(date, dm30 = date - 30)], on = .(date >= dm30, date < date), 
                 .N, by = .EACHI]$N][]

         date brandname status case number count
1: 2017-01-01        x1 closed       12345     0
2: 2017-01-01        x2 closed       12345     0
3: 2017-01-01        x3 closed       12345     0
4: 2017-01-02        x4   open        7864     3
5: 2017-01-03        x5   open       78642     4

解释

解决方案主要包括三个步骤：

DT[, .(date, dm30 = date - 30)]创建一个帮助表，其中包含要聚合的日期范围。
然后，使用 non-equi join 条件与DT正确连接，同时按连接条件进行聚合。 .N是一个特殊符号，包含每个组中的行数。
最后，计数作为新列添加到DF。

Answer 2

这是一个解决方案：它与问题略有不同。此版本使用数据框中的日期来选择另一个数据框中的子集。

对于日期范围中的每个日期，它计算另一个数据帧中的出现次数（每个日期和日期之间 - 30天）。然后，计数结果将存储在连续日期列表旁边的新列中。

数据：

-Dataframe- date 1 2016-07-12 2 2016-08-03 3 2016-08-24 4 2016-09-27 ... 265 2017-09-27 266 2017-09-28 267 2017-09-28 268 2017-09-28

<强>代码：

churn_frame <- as.data.frame(dates) #this is a sequential list of dates (calendar) churn_frame$new <- sapply(churn_frame$dates, function(x){ sum(dataframe$date <= x & dataframe$date >= x - 29)} #-29 days to cover exactly 30 days. )

<强>结果：

date new 1 2017-07-31 10 2 2017-08-01 10 3 2017-08-02 11 4 2017-08-03 10 5 2017-08-04 12 ... 58 2017-09-26 11 59 2017-09-27 12 60 2017-09-28 14 61 2017-09-29 12

在R中的日期之间运行行数

2 个答案:

解释