将 Pandas DataFrame 写入 Google Cloud Storage 或 BigQuery

Question

Hello and thanks for your time and consideration. I am developing a Jupyter Notebook in the Google Cloud Platform / Datalab. I have created a Pandas DataFrame and would like to write this DataFrame to both Google Cloud Storage(GCS) and/or BigQuery. I have a bucket in GCS and have, via the following code, created the following objects:

import gcp
import gcp.storage as storage
project = gcp.Context.default().project_id    
bucket_name = 'steve-temp'           
bucket_path  = bucket_name   
bucket = storage.Bucket(bucket_path)
bucket.exists()  

I have tried various approaches based on Google Datalab documentation but continue to fail. Thanks

Answer 1

Try the following working example:

from datalab.context import Context
import google.datalab.storage as storage
import google.datalab.bigquery as bq
import pandas as pd

# Dataframe to write
simple_dataframe = pd.DataFrame(data=[{1,2,3},{4,5,6}],columns=['a','b','c'])

sample_bucket_name = Context.default().project_id + '-datalab-example'
sample_bucket_path = 'gs://' + sample_bucket_name
sample_bucket_object = sample_bucket_path + '/Hello.txt'
bigquery_dataset_name = 'TestDataSet'
bigquery_table_name = 'TestTable'

# Define storage bucket
sample_bucket = storage.Bucket(sample_bucket_name)

# Create storage bucket if it does not exist
if not sample_bucket.exists():
    sample_bucket.create()

# Define BigQuery dataset and table
dataset = bq.Dataset(bigquery_dataset_name)
table = bq.Table(bigquery_dataset_name + '.' + bigquery_table_name)

# Create BigQuery dataset
if not dataset.exists():
    dataset.create()

# Create or overwrite the existing table if it exists
table_schema = bq.Schema.from_data(simple_dataframe)
table.create(schema = table_schema, overwrite = True)

# Write the DataFrame to GCS (Google Cloud Storage)
%storage write --variable simple_dataframe --object $sample_bucket_object

# Write the DataFrame to a BigQuery table
table.insert(simple_dataframe)

I used this example, and the _table.py file from the datalab github site as a reference. You can find other datalab source code files at this link.

Answer 2

Uploading to Google Cloud Storage without writing a temporary file and only using the standard GCS module

from google.cloud import storage
import os
import pandas as pd

# Only need this if you're running this code locally.
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = r'/your_GCP_creds/credentials.json'

df = pd.DataFrame(data=[{1,2,3},{4,5,6}],columns=['a','b','c'])

client = storage.Client()
bucket = client.get_bucket('my-bucket-name')
    
bucket.blob('upload_test/test.csv').upload_from_string(df.to_csv(), 'text/csv')

Answer 3

Using the Google Cloud Datalab documentation

import datalab.storage as gcs
gcs.Bucket('bucket-name').item('to/data.csv').write_to(simple_dataframe.to_csv(),'text/csv')

Answer 4

Writing a Pandas DataFrame to BigQuery

Update on @Anthonios Partheniou's answer.
The code is a bit different now - as of Nov. 29 2017

To define a BigQuery dataset

Pass a tuple containing project_id and dataset_id to bq.Dataset .

# define a BigQuery dataset    
bigquery_dataset_name = ('project_id', 'dataset_id')
dataset = bq.Dataset(name = bigquery_dataset_name)

To define a BigQuery table

Pass a tuple containing project_id , dataset_id and the table name to bq.Table .

# define a BigQuery table    
bigquery_table_name = ('project_id', 'dataset_id', 'table_name')
table = bq.Table(bigquery_table_name)

Create the dataset/ table and write to table in BQ

# Create BigQuery dataset
if not dataset.exists():
    dataset.create()

# Create or overwrite the existing table if it exists
table_schema = bq.Schema.from_data(dataFrame_name)
table.create(schema = table_schema, overwrite = True)

# Write the DataFrame to a BigQuery table
table.insert(dataFrame_name)

Answer 5

Since 2017, Pandas has a Dataframe to BigQuery function pandas.DataFrame.to_gbq

The documentation has an example:

import pandas_gbq as gbq gbq.to_gbq(df, 'my_dataset.my_table', projectid, if_exists='fail')

Parameter if_exists can be set to 'fail', 'replace' or 'append'

See also this example .

Answer 6

I spent a lot of time to find the easiest way to solve this:

import pandas as pd

df = pd.DataFrame(...)

df.to_csv('gs://bucket/path')

Answer 7

I have a little bit simpler solution for the task using Dask . You can convert your DataFrame to Dask DataFrame, which can be written to csv on Cloud Storage

import dask.dataframe as dd
import pandas
df # your Pandas DataFrame
ddf = dd.from_pandas(df,npartitions=1, sort=True)
dd.to_csv('gs://YOUR_BUCKET/ddf-*.csv', index=False, sep=',', header=False,  
                               storage_options={'token': gcs.session.credentials})  

Answer 8

我认为你需要将它加载到一个普通的字节变量中，并在一个单独的单元格中使用 %%storage write --variable $sample_bucketpath(see the doc) ......我仍在弄清楚......但这大致与读取 CSV 文件所需要做的相反，我不知道它是否对写入有影响，但我必须使用 BytesIO 来读取由 %% storage read 命令创建的缓冲区......希望它帮助，让我知道！

Answer 9

To Google storage :

def write_df_to_gs(df, gs_key):
    df.to_csv(gs_key)    

To BigQuery :

def upload_df_to_bq(df, project, bq_table):
    df.to_gbq(bq_table, project_id=project, if_exists='replace')

Answer 10

To save a parquet file in GCS with authentication due Service Account:

df.to_parquet("gs://<bucket-name>/file.parquet",
               storage_options={"token": <path-to-gcs-service-account-file>}