MaxKJC's picture
Upload folder using huggingface_hub
0a665df verified
Raw History Blame Contribute Delete
3.49 kB
# ============================================================
# Task: sales_flow_forecasting
# ============================================================
# Task objective:
# - Use historical sales of Store Sales store-category nodes to predict future sales.
# - This is a timestamp-level time series forecasting task across multiple store-family nodes.
#
# Core semantics:
# - sample is a decision timestamp.
# - inputs are the historical time series available before that timestamp, plus future covariates known at prediction time, such as onpromotion and is_holiday.
# - prediction is each episode's sales sequence over the future horizon.
# - Future sales (protocol time series ID: sales) must be excluded from inputs to prevent label leakage.
# ============================================================
task:
id: sales_flow_forecasting
type: time_series_forecasting
description: >
Predict future Store Sales flow from historical sales sequences for all
(store_nbr, family) episodes. Each sample corresponds to one decision
timestamp and uses index-reference labels.
inherit_general: true
sample:
primary_key: timestamp
label_mode: index_reference
inputs:
time_series:
- fields: sales
source: sales_flow_store_sales
window:
length: 28
unit: day
sampling:
native_rate: 1day
task_rate: 1day
downsample_method: mean
upsample_method: ffill
- fields: onpromotion
source: store_family_known_future_onpromotion
window:
length: 28
unit: day
sampling:
native_rate: 1day
task_rate: 1day
downsample_method: mean
upsample_method: ffill
- fields: is_holiday
source: store_family_known_future_holiday
window:
length: 28
unit: day
sampling:
native_rate: 1day
task_rate: 1day
downsample_method: mean
upsample_method: ffill
tabular_static:
source: store_sales_static_metadata
fields:
- store_nbr
- family
- family_id
- city
- state
- type
- cluster
graph:
source: store_family_relation_graph
prediction:
field: sales
source: sales_flow_store_sales
horizon:
length: 7
unit: day
starts_after: timestamp
stride: 7
label_storage:
ref_column: target_series_ref
shape_column: target_tensor_shape
leakage:
observed_past_only_columns:
- sales
known_future_columns:
- onpromotion
- is_holiday
data_refs:
time_series:
sales_flow_store_sales:
file: Store_Sales_protocol/input_multimodal/time_series/sales/sales.h5
value_field: sales
native_grid_minutes: 1440
store_family_known_future_onpromotion:
file: Store_Sales_protocol/input_multimodal/time_series/onpromotion/onpromotion.h5
value_field: onpromotion
native_grid_minutes: 1440
store_family_known_future_holiday:
file: Store_Sales_protocol/input_multimodal/time_series/is_holiday/is_holiday.h5
value_field: is_holiday
tabular_static:
store_sales_static_metadata:
file: Store_Sales_protocol/input_multimodal/meta_features.csv
id_column: episode_id
graph:
store_family_relation_graph:
file: Store_Sales_protocol/input_multimodal/node_level_task_graphs/store_family_relation_graph/edge_infos.csv