[timeseries] Speed up the train/val splitter

[shchur]

Description of changes:

• Speed up the append_suffix_to_item_id function
• Replace the surprisingly slow DataFrame.loc[index] operation with DataFrame.query("item_id in @index").

Testing on a subset of 5000 items from the M5 competition dataset:

Using code currently on master:

Loaded dataset with 7559974 rows and 5000 items.
df.slice_by_timestep(None, -prediction_length): 5.75s
LastWindowSplitter.split(df, prediction_length): 63.83s

After current PR:

Loaded dataset with 7559974 rows and 5000 items.
df.slice_by_timestep(None, -prediction_length): 1.38s
LastWindowSplitter.split(df, prediction_length): 8.30s

Code for reproducing the results

import time
import pandas as pd
from autogluon.timeseries import TimeSeriesDataFrame
from autogluon.timeseries.splitter import LastWindowSplitter

prediction_length = 28
# Dataset consists of the first 5000 items of the M5 competition dataset
raw_data = pd.read_parquet("../m5/data/subset.parquet")
static = pd.read_parquet("../m5/data/static.parquet")

raw_data["item_id"] = raw_data["item_id"].astype("str")
static["item_id"] = static["item_id"].astype("str")
static.set_index("item_id", inplace=True)

print(f"Loaded dataset with {len(raw_data)} rows and {raw_data['item_id'].nunique()} items.")
df = TimeSeriesDataFrame(raw_data, static_features=static)

start = time.time()
df.slice_by_timestep(None, -prediction_length)
print(f"df.slice_by_timestep(None, -prediction_length): {time.time() - start:.2f}s")

start = time.time()
splitter = LastWindowSplitter()
train_data, val_data = splitter.split(df, prediction_length)
print(f"LastWindowSplitter.split(df, prediction_length): {time.time() - start:.2f}s")

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[github-actions]

Job PR-2586-77c7f57 is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-2586/77c7f57/index.html

[canerturkmen]

LGTM! Just some questions.

[github-actions]

Job PR-2586-12d2b30 is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-2586/12d2b30/index.html

[github-actions]

Job PR-2586-7d41cb1 is done.
Docs are uploaded to http://autogluon-staging.s3-website-us-west-2.amazonaws.com/PR-2586/7d41cb1/index.html