Table / Pagination / H2O-3 Dataframe
Use a paginated table to display large (100m+ rows) tabular data using a H2O-3 dataframe.
import os
from time import time
import h2o
from h2o_wave import Q, app, main, ui
from loguru import logger
# This example requires H2O-3 to be running.
@app("/demo")
async def serve(q: Q):
logger.info(q.args)
logger.info(q.events)
if not q.app.initialized:
# This is called the first time our app runs
# Variables created here will be the same of all users of the app
# Save a direct link to our H2O Dataframe for all users to use throughout the app
try:
h2o.connect(url="http://127.0.0.1:54321")
except:
q.page['err'] = ui.form_card(box='1 1 4 2', items=[
ui.message_bar(type='error', text='Could not connect to H2O3. Please ensure H2O3 is running.'),
])
await q.page.save()
logger.error("H2O-3 is not running")
return
q.app.h2o_df = h2o.get_frame("py_6_sid_aff3")
# EXAMPLE OF CREATING A LARGE DATAFRAME
# h2o_df = h2o.create_frame(
# rows=1000000,
# cols=5,
# categorical_fraction=0.6,
# integer_fraction=0,
# binary_fraction=0,
# real_range=100,
# integer_range=100,
# missing_fraction=0,
# seed=1234,
# )
q.app.rows_per_page = 10 # TODO: How many rows do you want to show users at a time
# A list of booleans for if a column is sortable or not, by default
# we allow all and only numeric columns to be sorted based on H2O-3 functionality
# TODO: You may want to make a hardcoded list of [True, False] for your own use cases
q.app.column_sortable = q.app.h2o_df.isnumeric()
# A list of booleans for if a column is filterable or not, by default,
# we allow all and only categorical columns to be sorted based on H2O-3 functionality
# TODO: You may want to make a hardcoded list of [True, False] for your own use cases
q.app.column_filterable = q.app.h2o_df.isfactor()
# A list of booleans for if a column is searchable or not, by default,
# we allow all and only categorical and string columns to be sorted based on H2O-3 functionality
# TODO: You may want to make a hardcoded list of [True, False] for your own use cases
q.app.column_searchable = q.app.h2o_df.isfactor() + q.app.h2o_df.isstring()
q.app.initialized = True
if not q.client.initialized:
# This is called for each new browser that visits the app
# Multiple users can interact with the table at the same time without interrupting each other
# Users can make multiple changes to the table such as sorting and filtering
q.client.search = None
q.client.sort = None
q.client.filters = None
q.client.page_offset = 0
q.client.total_rows = len(q.app.h2o_df)
# Create the default UI for this user
q.page["meta"] = ui.meta_card(box="")
q.page["table_card"] = ui.form_card(
box="1 1 -1 -1",
items=[
ui.table(
name="h2o_table", # TODO: if you change this, you need to remember to update the serve function
columns=[
ui.table_column(
name=q.app.h2o_df.columns[i],
label=q.app.h2o_df.columns[i],
sortable=q.app.column_sortable[i],
filterable=q.app.column_filterable[i],
searchable=q.app.column_searchable[i],
)
for i in range(len(q.app.h2o_df.columns))
],
rows=get_table_rows(q),
resettable=True,
downloadable=True,
pagination=ui.table_pagination(
total_rows=q.client.total_rows,
rows_per_page=q.app.rows_per_page,
),
events=[
"page_change",
"sort",
"filter",
"search",
"reset",
"download",
],
)
],
)
q.client.initialized = True
# Check if user triggered any table action and save it to local state for allowing multiple
# actions to be performed on the data at the same time, e.g. sort the filtered data etc.
if q.events.h2o_table:
logger.info("table event occurred")
if q.events.h2o_table.page_change:
logger.info(f"table page change: {q.events.h2o_table.page_change}")
q.client.page_offset = q.events.h2o_table.page_change.get("offset", 0)
if q.events.h2o_table.sort:
logger.info(f"table sort: {q.events.h2o_table.sort}")
q.client.sort = q.events.h2o_table.sort
q.client.page_offset = 0
if q.events.h2o_table.filter:
logger.info(f"table filter: {q.events.h2o_table.filter}")
q.client.filters = q.events.h2o_table.filter
q.client.page_offset = 0
if q.events.h2o_table.search is not None:
logger.info(f"table search: {q.events.h2o_table.search}")
q.client.search = q.events.h2o_table.search
q.client.page_offset = 0
if q.events.h2o_table.download:
await download_h2o_table(q)
if q.events.h2o_table.reset:
logger.info("table reset")
q.client.search = None
q.client.sort = None
q.client.filters = None
q.client.page_offset = 0
q.client.total_rows = len(q.app.h2o_df)
# Update the rows in our UI
# TODO: if you change where your table is located, this needs updating
q.page["table_card"].items[0].table.rows = get_table_rows(q)
q.page["table_card"].items[0].table.pagination.total_rows = q.client.total_rows
await q.page.save()
def get_table_rows(q: Q):
logger.info(
f"Creating new table for rows: {q.client.page_offset} to {q.client.page_offset + q.app.rows_per_page}"
)
working_frame = prepare_h2o_data(q)
# Bring our limited UI rows locally to pandas to prepare for our ui.table
local_df = working_frame[
q.client.page_offset:q.client.page_offset + q.app.rows_per_page, :
].as_data_frame()
q.client.total_rows = len(working_frame)
table_rows = [
ui.table_row(
name=str(
q.client.page_offset + i
), # name is the index on the h2o dataframe for appropriate lookup
cells=[str(local_df[col].values[i]) for col in local_df.columns.to_list()],
)
for i in range(len(local_df))
]
h2o.remove(working_frame) # remove our duplicate work
return table_rows
async def download_h2o_table(q: Q):
# Create a unique file name as this is a multi-user app
local_file_path = f"h2o3_data_{str(int(time()))}.csv"
working_frame = prepare_h2o_data(q)
h2o.download_csv(working_frame, local_file_path)
(wave_file_path,) = await q.site.upload([local_file_path])
os.remove(local_file_path)
q.page["meta"].script = ui.inline_script(f'window.open("{wave_file_path}")')
def prepare_h2o_data(q: Q):
# This is used to prep the data we want to show on the screen or download, so it gets its own function
# If you have 5 users at the same time, there will be 6 large dataframes in h2o3 - ensure proper cluster size
working_frame = h2o.deep_copy(q.app.h2o_df, "working_df")
if q.client.sort is not None:
# H2O-3 can only sort numeric values - if the developer allows users to sort
# string columns the end users will see unexpected results
working_frame = working_frame.sort(
by=list(q.client.sort.keys()), ascending=list(q.client.sort.values())
)
if q.client.filters is not None:
for key in q.client.filters.keys():
working_frame = working_frame[
working_frame[key].match(q.client.filters[key])
]
if q.client.search is not None:
# We check if our search term is in any of the searchable columns
# Start with and index of 0s and then filter to only keep rows with index > 0
index = h2o.create_frame(
rows=len(working_frame), cols=1, integer_fraction=1, integer_range=1
)
index["C1"] = 0
for i in range(len(q.app.h2o_df.columns)):
if q.app.column_searchable[i]:
index = index + working_frame[q.app.h2o_df.columns[i]].grep(
pattern=q.client.search, ignore_case=True, output_logical=True
)
working_frame = working_frame[index]
return working_frame
Tags: form h2o3 pagination table