We Keep Coding

Nested Json Using pyspark

We have to build nested json using below structure in pyspark and i have added data that need to feed using this
Input Data structure
Data
from pyspark.sql import SparkSession
spark = SparkSession.builder.appName('sparkdf').getOrCreate()
a1=["DA_STinf","DA_Stinf_NA","DA_Stinf_city","DA_Stinf_NA_ID","DA_Stinf_NA_ID_GRANT","DA_country"]
a2=["data.studentinfo","data.studentinfo.name","data.studentinfo.city","data.studentinfo.name.id","data.studentinfo.name.id.grant","data.country"]
columns = ["data","action"]
df = spark.createDataFrame(zip(a1, a2), columns)
#Input data for json structure
a1=["Pune"]
a2=["YES"]
a3=["India"]
col=["DA_Stinf_city","DA_Stinf_NA_ID_GRANT","DA_country"]
data=spark.createDataFrame(zip(a1, a2,a3), col)
Expected result based on above data
{
"data": {
"studentinfo": {
"city": "Pune",
"name": {
"id": {
"grant": "YES"
}
}
},
"country": "india"
}
}
we have tried using F.struct function in manually but we have find dynamic way to build this json using df dataframe having data and action column
data.select(
F.struct(
F.struct(
F.struct(F.col("DA_Stinf_city")).alias("city"),
F.struct(
F.struct(F.col("DA_Stinf_NA_ID_GRANT")).alias("id")
).alias("name"),
).alias("studentinfo"),
F.struct(F.col("DA_country")).alias("country")
).alias("data")
)

The approach below should give the correct structure (with the wrong key names - if you are happy with the approach, which doesn't use DataFrame operations but rather works in the underlying RDD, then I can flesh it out):
def build_json(input, running={}):
new_input = {}
for hierarchy, value in input:
key = hierarchy.pop(0)
if len(hierarchy) == 0:
running[key] = value
else:
new_input[key] = new_input.get(key, []) + [(hierarchy, value)]
for key in new_input:
print(new_input[key])
running[key] = build_json(new_input[key], running={})
return running
data.rdd.map(
lambda x: build_json(
[(column.split("_"), value) for column, value in x.asDict().items()]
)
)
The basic idea is to get a set of tuples from the underlying RDD consisting of the column name broken into its json hierarchy and the value to insert into the hierarchy. Then the function build_json inserts the value into its correct place in the json hierarchy, while building out the json object recursively.

generating a Chart.js chart with python data

I am working on a litte stock-market project in Python.
Every week, a status.xlsx file is generated that tells me what sectors make up my portfolio, e.g.
xls = pd.ExcelFile('Status_week_9.xlsx')
df1=pd.read_excel(xls)
print(df1)
I am looping over all files, so I get a dataframe similar to this for every file:
sector pct
Consumer Cyclical 0.319638
Industrials 0.203268
Financial Services 0.200217
...
Now I would like to loop through these weekly reports and pass the data to a Chart.js template in order to get a stacked bar for each week that shows the increase / decrease of sectors over time, e.g. https://jsfiddle.net/qg1ya5vk/4/ (demo only, does not add up to 1)
The template file looks like this:
The idea was to use a template with placeholders for the chart and use
from string import Template
to replace the placeholders with the corresponding values.
The variables are "labels" and "dataset". "Labels" consists of the calendar weeks. This can be extracted from the filenames. So far, so good ;-)
Now for the "dataset", I'm not sure how to get information from the n dataframes.
One dataset would have to look like this:
{label: 'Energy',
data: [0.037975, 0.038512, 0.039817, 0.065010],}
So this would mean that the Energy sector had a share of 3.7975% in week 1, 3.8512% in week 2 etc. Complicating things even further, it's also possible that one sector is not present in one week, so I would have to add a 0 then, because the sector is not part of the original dataframe.
There can be n (comma-separated) datasets.
I probably have to transpose the dataframe but it doesn't quite do the trick (yet).
So far my code looks like this:
import pandas as pd
import glob
import os
from string import Template
labels=[]
dataset=[]
files =...
for i in files:
cw = i.split('_')[3].split('.')[0] ## extract calendar week from filename
xls = pd.ExcelFile(i)
df1 = pd.read_excel(xls, 'Sectors') ## select sheet 'Sectors'
df1['CW'] = cw ## add cw to df1
df1_t = df1.T. ## transpose df1
sectors = df1.sector.to_list()
share = df1.pct.to_list()
labels.append(cw)
dataset.append(df1_t) ##??
# {
# label: 'Energy',
# data: [0.037975, 0.038512, 0.039817, 0.065010],
# }
d = {'label' : labels, 'datasets' : dataset}
## open Chart template and put in values:
with open('template.txt', 'r') as f:
src = Template(f.read())
result = src.substitute(d)
print(result)
How would you generate the datasets?
I am also thinking this is a little bit long-winded and error-prone. Maybe there is another way to tackle this?
Template for chart:
var label = $label;
var ctx = document.getElementById("myChart4").getContext('2d');
var myChart = new Chart(ctx, {
type: 'bar',
data: {
labels: label,
datasets: [
$datasets //**this is the culprit**
],
},
options: {
tooltips: {
displayColors: true,
callbacks:{
mode: 'x',
},
},
scales: {
xAxes: [{
stacked: true,
gridLines: {
display: false,
}
}],
yAxes: [{
stacked: true,
ticks: {
beginAtZero: true,
},
type: 'linear',
}]
},
responsive: true,
maintainAspectRatio: false,
legend: { position: 'bottom' },
}
});

What you are looking for is the json module. You can simply arrange your data in the correct form in python and write to a JSON that Chart.js can understand.
import json
data = {'label' : labels, 'datasets' : dataset}
with open('data.json', 'w') as f:
json.dump(data, f)
This question is about how to get your JSON into Chart.js.

Convert two CSV tables with one-to-many relation to JSON with embedded list of subdocuments

I have two CSV files which have one-to-many relation between them.
main.csv:
"main_id","name"
"1","foobar"
attributes.csv:
"id","main_id","name","value","updated_at"
"100","1","color","red","2020-10-10"
"101","1","shape","square","2020-10-10"
"102","1","size","small","2020-10-10"
I would like to convert this to JSON of this structure:
[
{
"main_id": "1",
"name": "foobar",
"attributes": [
{
"id": "100",
"name": "color",
"value": "red",
"updated_at": "2020-10-10"
},
{
"id": "101",
"name": "shape",
"value": "square",
"updated_at": "2020-10-10"
},
{
"id": "103",
"name": "size",
"value": "small",
"updated_at": "2020-10-10"
}
]
}
]
I tried using Python and Pandas like:
import pandas
def transform_group(group):
group.reset_index(inplace=True)
group.drop('main_id', axis='columns', inplace=True)
return group.to_dict(orient='records')
main = pandas.read_csv('main.csv')
attributes = pandas.read_csv('attributes.csv', index_col=0)
attributes = attributes.groupby('main_id').apply(transform_group)
attributes.name = "attributes"
main = main.merge(
right=attributes,
on='main_id',
how='left',
validate='m:1',
copy=False,
)
main.to_json('out.json', orient='records', indent=2)
It works. But the issue is that it does not seem to scale. When running on my whole dataset I have, I can load individual CSV files without problems, but when trying to modify data structure before calling to_json, memory usage explodes.
So is there a more efficient way to do this transformation? Maybe there is some Pandas feature I am missing? Or is there some other library to use? Moreover, use of apply seems to be pretty slow here.

This is a tough problem and we have all felt your pain.
There are three ways I would attack this problem. First, groupby is slower if you allow pandas to do the break out.
import pandas as pd
import numpy as np
from collections import defaultdict
df = pd.DataFrame({'id': np.random.randint(0, 100, 5000),
'name': np.random.randint(0, 100, 5000)})
now if you do the standard groupby
groups = []
for k, rows in df.groupby('id'):
groups.append(rows)
you will find that
groups = defaultdict(lambda: [])
for id, name in df.values:
groups[id].append((id, name))
is about 3 times faster.
The second method is I would use change it to use Dask and the dask parallelization. A discussion about dask is what is dask and how is it different from pandas.
The third is algorithmic. Load up the main file and then by ID, then only load the data for that ID, having multiple bites at what is in memory and what is in disk, then saving out a partial result as it becomes available.

So in my case I was able to load original tables in memory, but doing embedding exploded the size so that it did not fit memory anymore. So I ended up still using Pandas to load CSV files, but then I iteratively generate row by row and saving each row into a separate JSON. This means I do not have a large data structure in the memory for one large JSON.
Another important realization was that it is important to make the related column an index, and that it has to be sorted, so that querying it is fast (because generally there are duplicate entries in the related column).
I made the following two helper functions:
def get_related_dict(related_table, label):
assert related_table.index.is_unique
if pandas.isna(label):
return None
row = related_table.loc[label]
assert isinstance(row, pandas.Series), label
result = row.to_dict()
result[related_table.index.name] = label
return result
def get_related_list(related_table, label):
# Important to be more performant when selecting non-unique labels.
assert related_table.index.is_monotonic_increasing
try:
# We use this syntax for always get a DataFrame and not a Series when there is only one row matching.
return related_table.loc[[label], :].to_dict(orient='records')
except KeyError:
return []
And then I do:
main = pandas.read_csv('main.csv', index_col=0)
attributes = pandas.read_csv('attributes.csv', index_col=1)
# We sort index to be more performant when selecting non-unique labels. We use stable sort.
attributes.sort_index(inplace=True, kind='mergesort')
columns = [main.index.name] + list(main.columns)
for row in main.itertuples(index=True, name=None):
assert len(columns) == len(row)
data = dict(zip(columns, row))
data['attributes'] = get_related_list(attributes, data['main_id'])
json.dump(data, sys.stdout, indent=2)
sys.stdout.write("\n")

Pandas json_normalize and JSON flattening error

A panda newbie here that's struggling to understand why I'm unable to completely flatten a JSON I receive from an API. I need a Dataframe with all the data that is returned by the API, however I need all nested data to be expanded and given it's own columns for me to be able to use it.
The JSON I receive is as follows:
[
{
"query":{
"id":"1596487766859-3594dfce3973bc19",
"name":"test"
},
"webPage":{
"inLanguages":[
{
"code":"en"
}
]
},
"product":{
"name":"Test",
"description":"Test2",
"mainImage":"image1.jpg",
"images":[
"image2.jpg",
"image3.jpg"
],
"offers":[
{
"price":"45.0",
"currency":"€"
}
],
"probability":0.9552192
}
}
]
Running pd.json_normalize(data) without any additional parameters shows the nested values price and currency in the product.offers column. When I try to separate these out into their own columns with the following:
pd.json_normalize(data,record_path=['product',meta['product',['offers']]])
I end up with the following error:
f"{js} has non list value {result} for path {spec}. "
Any help would be much appreciated.

I've used this technique a few times
do initial pd.json_normalize() to discover the columns
build meta parameter by inspecting this and the original JSON. NB possible index out of range here
you can only request one list drives record_path param
a few tricks product/images is a list so it gets named 0. rename it
did a Cartesian product to merge two different data frames from breaking down lists. It's not so stable
data = [{'query': {'id': '1596487766859-3594dfce3973bc19', 'name': 'test'},
'webPage': {'inLanguages': [{'code': 'en'}]},
'product': {'name': 'Test',
'description': 'Test2',
'mainImage': 'image1.jpg',
'images': ['image2.jpg', 'image3.jpg'],
'offers': [{'price': '45.0', 'currency': '€'}],
'probability': 0.9552192}}]
# build default to get column names
df = pd.json_normalize(data)
# from column names build the list that gets sent to meta param
mymeta = [[s for s in c.split(".")] for c in df.columns ]
# exclude lists from meta - this will fail
mymeta = [l for l in mymeta if not isinstance(data[0][l[0]][l[1]], list)]
# you can build df from either of the product lists NOT both
df1 = pd.json_normalize(data, record_path=[["product","offers"]], meta=mymeta)
df2 = pd.json_normalize(data, record_path=[["product","images"]], meta=mymeta).rename(columns={0:"image"})
# want them together - you can merge them. note columns heavily overlap so remove most columns from df2
df1.assign(foo=1).merge(
df2.assign(foo=1).drop(columns=[c for c in df2.columns if c!="image"]), on="foo").drop(columns="foo")

Merge Multiple CSV with different column name but same definition

I have different sources(CSV) of similar data set which i want to merge into single data and write it to my DB. Since data is coming from different sources, they use different headers in their CSV, i want to merge these columns with logical meaning.
So far, i have tried reading all headers first and re reading the files to first get all the data in a single data frame and then doing if else to merge the columns together with same meaning. Ideally I would like to create a mapping file with all possible column names per column and then read CSV using that mapping. The data is not ordered or sorted between files. Number of columns might be different too but they all have the columns i am interested in.
Sample data:
File 1:
id, name, total_amount...
1, "test", 123 ..
File 2:
member_id, tot_amnt, name
2, "test2", 1234 ..
i want this to look like
id, name, total_amount...
1, "test", 123...
2, "test2", 1234...
...
I can't think of an elegant way to do this, would be great to get some direction or help with this.
Thanks

Use skiprows and header=None to skip the header, names to specify your own list of column names, and concat to merge into a single df. i.e.
import pandas as pd
pd.concat([
pd.read_csv('file1.csv',skiprows=1,header=None,names=['a','b','c']),
pd.read_csv('file2.csv',skiprows=1,header=None,names=['a','b','c'])]
)
Edit: If the different files differ only by column order you can specify different column orders to names and if you want to select a subset of columns use usecols. But you need to do this mapping in advance, either by probing the file, or some other rule.
This requires mapping files to handlers somehow
i.e.
file1.csv
id, name, total_amount
1, "test", 123
file2.csv
member_id, tot_amnt, ignore, name
2, 1234, -1, "test2"
The following selects the common 3 columns and renames / reorders.
import pandas as pd
pd.concat([
pd.read_csv('file1.csv',skiprows=1,header=None,names=['id','name','value'],usecols=[0,1,2]),
pd.read_csv('file2.csv',skiprows=1,header=None,names=['id','value','name'],usecols=[0,1,3])],
sort=False
)
Edit 2:
And a nice way to apply this is to use lambda's and maps - i.e.
parsers = {
"schema1": lambda f: pd.read_csv(f,skiprows=1,header=None,names=['id','name','value'],usecols=[0,1,2]),
"schema2": lambda f: pd.read_csv(f,skiprows=1,header=None,names=['id','value','name'],usecols=[0,1,3])
}
map = {
"file2.csv": "schema2",
"file1.csv": "schema1"}
pd.concat([parsers[v](k) for k,v in map.items()], sort=False)

This is what i ended up doing and found to be the cleanest solution. Thanks David your help.
dict1= {'member_number': 'id', 'full name': 'name', …}
dict2= {'member_id': 'id', 'name': 'name', …}
parsers = {
"schema1": lambda f, dict: pd.read_csv(f,index_col=False,usecols=list(dict.keys())),
"schema2": lambda f, dict: pd.read_csv(f,index_col=False,usecols=list(dict.keys()))
}
map = {
'schema1': (a_file.csv,dict1),
'schema2': (b_file.csv,dict2)
}
total = []
for k,v in map.items():
d = parsers[k](v[0], v[1])
d.rename(columns=v[1], inplace=True)
total.append(d)
final_df = pd.concat(total, sort=False)