XFrames: Another Convenient Python Interface to Spark

Currently, pyspark might be the most popular python interface to Apache Spark. However, the xframes package ( https://github.com/cchayden/xframes ) definitely is an alternative worth trying.

As shown in the code snippet below, the XFrame, which is the dataframe object in the xframes package, interacts well with other python data structures and numpy functions. To me, the XFrame is easier to work with than the pyspark.dataframe and has more “authentic” python flavor.

from xframes import XFrame, aggregate

df = XFrame.read_csv("Downloads/nycflights.csv", header = True, nrows = 11)

### SUBSETTING
sel_cols = ["origin", "dest", "distance", "dep_delay", "carrier"]

df2 = df[sel_cols]
# OR:
# df.sql("select " + ", ".join(sel_cols) + " from df")

### FILTERING ###
print df2[(df2["origin"] == 'EWR') & (df2["carrier"] == "UA")]
# OR:
# print df2.filterby("EWR", "origin").filterby("UA", "carrier")

### AGGREGATING ###
from numpy import median

grp1 = df2.groupby("origin", {"dist": aggregate.CONCAT("distance")})

agg1 = XFrame({"origin": grp1["origin"], "med_dist": map(median, grp1["dist"])})
# OR:
# grp1["med_dist"] = grp1.apply(lambda row: median(row["dist"]))
# agg1 = grp1[["origin", "med_dist"]]
# USING SQL:
# df2.sql("select origin, percentile_approx(distance, 0.5) as med_dist from df2 group by origin")

for row in agg1:
  print row
# {'origin': u'LGA', 'med_dist': 747.5}
# {'origin': u'JFK', 'med_dist': 1089.0}
# {'origin': u'EWR', 'med_dist': 1065.0}

agg2 = df2.groupby("origin", {"avg_delay": aggregate.MEAN("dep_delay")})
# USING SQL:
# df2.sql("select origin, mean(dep_delay) as avg_delay from df2 group by origin")

for row in agg2:
  print row
# {'origin': u'LGA', 'avg_delay': -1.75}
# {'origin': u'JFK', 'avg_delay': -0.6666666666666666}
# {'origin': u'EWR', 'avg_delay': -2.3333333333333335}

### JOINING ###
for row in  agg1.join(agg2, on = {"origin": "origin"}, how = "inner"):
    print row
# {'origin': u'LGA', 'med_dist': 747.5, 'avg_delay': -1.75}
# {'origin': u'JFK', 'med_dist': 1089.0, 'avg_delay': -0.6666666666666666}
# {'origin': u'EWR', 'med_dist': 1065.0, 'avg_delay': -2.3333333333333335}