如何将 Cassandra Map 转换为 Pandas Dataframe

Question

我想从 map<string, int> 类型的 cassandra 列族中读取数据，并将其转换为 Pandas 数据框。我还想用它来训练python中的模型，正如在虹膜物种分类中提到的here。

如果，我会使用 csv 来训练模型。然后它看起来像这样：

label,  f1, f2, f3, f4, f5
  0  ,  11 , 1, 6 , 1,  2  
  1  ,  5,   5, 1 , 2,  6
  0  ,  12,  9, 3 , 6,  8
  0  ,  9,  3,  8,  1,  0 

Cassandra 列族：

                  FeatureSet                    |   label

{'f1': 11, 'f2': 1, 'f3': 6, 'f4': 1, 'f5': 2}  |     0
{'f1': 5, 'f2':  5, 'f3': 1, 'f4': 2, 'f5': 6}  |     1
{'f1': 12, 'f2': 9, 'f3': 3, 'f4': 6, 'f5': 8}  |     0
{'f1': 9, 'f2': 3, 'f3': 8, 'f4': 1, 'f5': 0}   |     0

代码：

import pandas as pd
from sklearn2pmml import PMMLPipeline
from sklearn.tree import DecisionTreeClassifier
from cassandra.cluster import Cluster

CASSANDRA_HOST = ['172.16.X.Y','172.16.X1.Y1'] 
CASSANDRA_PORT = 9042
CASSANDRA_DB = "KEYSPACE"
CASSANDRA_TABLE = "COLUMNFAMILY"

cluster = Cluster(contact_points=CASSANDRA_HOST, port=CASSANDRA_PORT)
session = cluster.connect(CASSANDRA_DB)

sql_query = "SELECT * FROM {}.{};".format(CASSANDRA_DB, CASSANDRA_TABLE)

df = pd.DataFrame()

for row in session.execute(sql_query):  
            What should i write here and get X_train, Y_train in pandas dataframe 



iris_pipeline = PMMLPipeline([
    ("classifier", DecisionTreeClassifier())
])
iris_pipeline.fit(X_train, Y_train)

Answer 1

除了 MaxU 答案之外，如果您想将结果视为数据框，您需要做的就是再添加一行：

df = pd.DataFrame(rslt._current_rows)

Answer 2

我针对同一问题发布了一个有效的解决方案here，以将OrderedMapSerializedKey Cassandra 映射字段作为字典读取到您的数据框中。

---

编辑：

在之前的解决方案中，我只替换了 Cassandra 数据集的第一（0）行（rows 是元组列表，其中每个元组都是 Cassandra 中的一行）

from cassandra.util import OrderedMapSerializedKey

def pandas_factory(colnames, rows):

    # Convert tuple items of 'rows' into list (elements of tuples cannot be replaced)
    rows = [list(i) for i in rows]

    # Convert only 'OrderedMapSerializedKey' type list elements into dict
    for idx_row, i_row in enumerate(rows):

        for idx_value, i_value in enumerate(i_row):

            if type(i_value) is OrderedMapSerializedKey:

                rows[idx_row][idx_value] = dict(rows[idx_row][idx_value])

    return pd.DataFrame(rows, columns=colnames)

Answer 3

你可以使用this approach:

import pandas as pd
from cassandra.cluster import Cluster

def pandas_factory(colnames, rows):
    return pd.DataFrame(rows, columns=colnames)

CASSANDRA_HOST = ['172.16.X.Y','172.16.X1.Y1'] 
CASSANDRA_PORT = 9042
CASSANDRA_DB = "KEYSPACE"
CASSANDRA_TABLE = "COLUMNFAMILY"

cluster = Cluster(contact_points=CASSANDRA_HOST, port=CASSANDRA_PORT)
session = cluster.connect(CASSANDRA_DB)

session.row_factory = pandas_factory
session.default_fetch_size = None

query = "SELECT * FROM {}.{};".format(CASSANDRA_DB, CASSANDRA_TABLE)

rslt = session.execute(query, timeout=None)
df = rslt._current_rows