Refactor review pipeline around normalized items

This commit is contained in:
ben
2026-03-20 11:27:46 -04:00
parent 607c51038a
commit 9104781b93
6 changed files with 512 additions and 361 deletions

View File

@@ -3,11 +3,8 @@ from pathlib import Path
import click
import build_canonical_layer
import build_observed_products
import validate_cross_retailer_flow
from enrich_giant import format_decimal, to_decimal
from layer_helpers import read_csv_rows, stable_id, write_csv_rows
from layer_helpers import read_csv_rows, write_csv_rows
PURCHASE_FIELDS = [
@@ -15,13 +12,18 @@ PURCHASE_FIELDS = [
"retailer",
"order_id",
"line_no",
"observed_item_key",
"observed_product_id",
"canonical_product_id",
"normalized_row_id",
"normalized_item_id",
"catalog_id",
"review_status",
"resolution_action",
"raw_item_name",
"normalized_item_name",
"catalog_name",
"category",
"product_type",
"brand",
"variant",
"image_url",
"retailer_item_id",
"upc",
@@ -55,7 +57,7 @@ PURCHASE_FIELDS = [
EXAMPLE_FIELDS = [
"example_name",
"canonical_product_id",
"catalog_id",
"giant_purchase_date",
"giant_raw_item_name",
"giant_price_per_lb",
@@ -66,8 +68,8 @@ EXAMPLE_FIELDS = [
]
CATALOG_FIELDS = [
"canonical_product_id",
"canonical_name",
"catalog_id",
"catalog_name",
"category",
"product_type",
"brand",
@@ -81,9 +83,20 @@ CATALOG_FIELDS = [
"updated_at",
]
PRODUCT_LINK_FIELDS = [
"normalized_item_id",
"catalog_id",
"link_method",
"link_confidence",
"review_status",
"reviewed_by",
"reviewed_at",
"link_notes",
]
RESOLUTION_FIELDS = [
"observed_product_id",
"canonical_product_id",
"normalized_item_id",
"catalog_id",
"resolution_action",
"status",
"resolution_notes",
@@ -91,10 +104,6 @@ RESOLUTION_FIELDS = [
]
def decimal_or_zero(value):
return to_decimal(value) or Decimal("0")
def derive_metrics(row):
line_total = to_decimal(row.get("net_line_total") or row.get("line_total"))
qty = to_decimal(row.get("qty"))
@@ -162,10 +171,7 @@ def derive_metrics(row):
def order_lookup(rows, retailer):
return {
(retailer, row["order_id"]): row
for row in rows
}
return {(retailer, row["order_id"]): row for row in rows}
def read_optional_csv_rows(path):
@@ -175,28 +181,10 @@ def read_optional_csv_rows(path):
return read_csv_rows(path)
def load_resolution_lookup(resolution_rows):
lookup = {}
for row in resolution_rows:
if not row.get("observed_product_id"):
continue
lookup[row["observed_product_id"]] = row
return lookup
def merge_catalog_rows(existing_rows, auto_rows):
merged = {}
for row in auto_rows + existing_rows:
canonical_product_id = row.get("canonical_product_id", "")
if canonical_product_id:
merged[canonical_product_id] = row
return sorted(merged.values(), key=lambda row: row["canonical_product_id"])
def catalog_row_from_canonical(row):
def normalize_catalog_row(row):
return {
"canonical_product_id": row.get("canonical_product_id", ""),
"canonical_name": row.get("canonical_name", ""),
"catalog_id": row.get("catalog_id") or row.get("canonical_product_id", ""),
"catalog_name": row.get("catalog_name") or row.get("canonical_name", ""),
"category": row.get("category", ""),
"product_type": row.get("product_type", ""),
"brand": row.get("brand", ""),
@@ -211,24 +199,67 @@ def catalog_row_from_canonical(row):
}
def build_link_state(enriched_rows):
observed_rows = build_observed_products.build_observed_products(enriched_rows)
canonical_rows, link_rows = build_canonical_layer.build_canonical_layer(observed_rows)
giant_row, costco_row = validate_cross_retailer_flow.find_proof_pair(observed_rows)
canonical_rows, link_rows, _proof_rows = validate_cross_retailer_flow.merge_proof_pair(
canonical_rows,
link_rows,
giant_row,
costco_row,
)
def is_review_first_catalog_row(row):
notes = row.get("notes", "").strip().lower()
if notes.startswith("auto-linked via"):
return False
return True
observed_id_by_key = {
row["observed_key"]: row["observed_product_id"] for row in observed_rows
def normalize_link_row(row):
return {
"normalized_item_id": row.get("normalized_item_id", ""),
"catalog_id": row.get("catalog_id") or row.get("canonical_product_id", ""),
"link_method": row.get("link_method", ""),
"link_confidence": row.get("link_confidence", ""),
"review_status": row.get("review_status", ""),
"reviewed_by": row.get("reviewed_by", ""),
"reviewed_at": row.get("reviewed_at", ""),
"link_notes": row.get("link_notes", ""),
}
canonical_id_by_observed = {
row["observed_product_id"]: row["canonical_product_id"] for row in link_rows
def normalize_resolution_row(row):
return {
"normalized_item_id": row.get("normalized_item_id", ""),
"catalog_id": row.get("catalog_id") or row.get("canonical_product_id", ""),
"resolution_action": row.get("resolution_action", ""),
"status": row.get("status", ""),
"resolution_notes": row.get("resolution_notes", ""),
"reviewed_at": row.get("reviewed_at", ""),
}
return observed_rows, canonical_rows, link_rows, observed_id_by_key, canonical_id_by_observed
def load_resolution_lookup(resolution_rows):
lookup = {}
for row in resolution_rows:
normalized_row = normalize_resolution_row(row)
normalized_item_id = normalized_row.get("normalized_item_id", "")
if not normalized_item_id:
continue
lookup[normalized_item_id] = normalized_row
return lookup
def merge_catalog_rows(existing_rows, new_rows):
merged = {}
for row in existing_rows + new_rows:
normalized_row = normalize_catalog_row(row)
catalog_id = normalized_row.get("catalog_id", "")
if catalog_id:
merged[catalog_id] = normalized_row
return sorted(merged.values(), key=lambda row: row["catalog_id"])
def load_link_lookup(link_rows):
lookup = {}
for row in link_rows:
normalized_row = normalize_link_row(row)
normalized_item_id = normalized_row.get("normalized_item_id", "")
if not normalized_item_id:
continue
lookup[normalized_item_id] = normalized_row
return lookup
def build_purchase_rows(
@@ -237,25 +268,37 @@ def build_purchase_rows(
giant_orders,
costco_orders,
resolution_rows,
link_rows=None,
catalog_rows=None,
):
all_enriched_rows = giant_enriched_rows + costco_enriched_rows
(
observed_rows,
canonical_rows,
link_rows,
observed_id_by_key,
canonical_id_by_observed,
) = build_link_state(all_enriched_rows)
resolution_lookup = load_resolution_lookup(resolution_rows)
for observed_product_id, resolution in resolution_lookup.items():
link_lookup = load_link_lookup(link_rows or [])
catalog_lookup = {
row["catalog_id"]: normalize_catalog_row(row)
for row in (catalog_rows or [])
if normalize_catalog_row(row).get("catalog_id")
}
for normalized_item_id, resolution in resolution_lookup.items():
action = resolution.get("resolution_action", "")
status = resolution.get("status", "")
if status != "approved":
continue
if action in {"link", "create"} and resolution.get("canonical_product_id"):
canonical_id_by_observed[observed_product_id] = resolution["canonical_product_id"]
if action in {"link", "create"} and resolution.get("catalog_id"):
link_lookup[normalized_item_id] = {
"normalized_item_id": normalized_item_id,
"catalog_id": resolution["catalog_id"],
"link_method": f"manual_{action}",
"link_confidence": "high",
"review_status": status,
"reviewed_by": "",
"reviewed_at": resolution.get("reviewed_at", ""),
"link_notes": resolution.get("resolution_notes", ""),
}
elif action == "exclude":
canonical_id_by_observed[observed_product_id] = ""
link_lookup.pop(normalized_item_id, None)
orders_by_id = {}
orders_by_id.update(order_lookup(giant_orders, "giant"))
orders_by_id.update(order_lookup(costco_orders, "costco"))
@@ -265,24 +308,30 @@ def build_purchase_rows(
all_enriched_rows,
key=lambda item: (item["order_date"], item["retailer"], item["order_id"], int(item["line_no"])),
):
observed_key = build_observed_products.build_observed_key(row)
observed_product_id = observed_id_by_key.get(observed_key, "")
normalized_item_id = row.get("normalized_item_id", "")
resolution = resolution_lookup.get(normalized_item_id, {})
link_row = link_lookup.get(normalized_item_id, {})
catalog_row = catalog_lookup.get(link_row.get("catalog_id", ""), {})
order_row = orders_by_id.get((row["retailer"], row["order_id"]), {})
metrics = derive_metrics(row)
resolution = resolution_lookup.get(observed_product_id, {})
purchase_rows.append(
{
"purchase_date": row["order_date"],
"retailer": row["retailer"],
"order_id": row["order_id"],
"line_no": row["line_no"],
"observed_item_key": row["observed_item_key"],
"observed_product_id": observed_product_id,
"canonical_product_id": canonical_id_by_observed.get(observed_product_id, ""),
"normalized_row_id": row.get("normalized_row_id", ""),
"normalized_item_id": normalized_item_id,
"catalog_id": link_row.get("catalog_id", ""),
"review_status": resolution.get("status", ""),
"resolution_action": resolution.get("resolution_action", ""),
"raw_item_name": row["item_name"],
"normalized_item_name": row["item_name_norm"],
"catalog_name": catalog_row.get("catalog_name", ""),
"category": catalog_row.get("category", ""),
"product_type": catalog_row.get("product_type", ""),
"brand": catalog_row.get("brand", ""),
"variant": catalog_row.get("variant", ""),
"image_url": row.get("image_url", ""),
"retailer_item_id": row["retailer_item_id"],
"upc": row["upc"],
@@ -307,33 +356,7 @@ def build_purchase_rows(
**metrics,
}
)
return purchase_rows, observed_rows, canonical_rows, link_rows
def apply_manual_resolutions_to_links(link_rows, resolution_rows):
link_by_observed = {row["observed_product_id"]: dict(row) for row in link_rows}
for resolution in resolution_rows:
if resolution.get("status") != "approved":
continue
observed_product_id = resolution.get("observed_product_id", "")
action = resolution.get("resolution_action", "")
if not observed_product_id:
continue
if action == "exclude":
link_by_observed.pop(observed_product_id, None)
continue
if action in {"link", "create"} and resolution.get("canonical_product_id"):
link_by_observed[observed_product_id] = {
"observed_product_id": observed_product_id,
"canonical_product_id": resolution["canonical_product_id"],
"link_method": f"manual_{action}",
"link_confidence": "high",
"review_status": resolution.get("status", ""),
"reviewed_by": "",
"reviewed_at": resolution.get("reviewed_at", ""),
"link_notes": resolution.get("resolution_notes", ""),
}
return sorted(link_by_observed.values(), key=lambda row: row["observed_product_id"])
return purchase_rows, sorted(link_lookup.values(), key=lambda row: row["normalized_item_id"])
def build_comparison_examples(purchase_rows):
@@ -342,7 +365,7 @@ def build_comparison_examples(purchase_rows):
for row in purchase_rows:
if row.get("normalized_item_name") != "BANANA":
continue
if not row.get("canonical_product_id"):
if not row.get("catalog_id"):
continue
if row["retailer"] == "giant" and row.get("price_per_lb"):
giant_banana = row
@@ -355,7 +378,7 @@ def build_comparison_examples(purchase_rows):
return [
{
"example_name": "banana_price_per_lb",
"canonical_product_id": giant_banana["canonical_product_id"],
"catalog_id": giant_banana["catalog_id"],
"giant_purchase_date": giant_banana["purchase_date"],
"giant_raw_item_name": giant_banana["raw_item_name"],
"giant_price_per_lb": giant_banana["price_per_lb"],
@@ -389,27 +412,29 @@ def main(
examples_csv,
):
resolution_rows = read_optional_csv_rows(resolutions_csv)
purchase_rows, _observed_rows, canonical_rows, link_rows = build_purchase_rows(
catalog_rows = merge_catalog_rows(
[row for row in read_optional_csv_rows(catalog_csv) if is_review_first_catalog_row(row)],
[],
)
existing_links = [normalize_link_row(row) for row in read_optional_csv_rows(links_csv)]
purchase_rows, link_rows = build_purchase_rows(
read_csv_rows(giant_items_enriched_csv),
read_csv_rows(costco_items_enriched_csv),
read_csv_rows(giant_orders_csv),
read_csv_rows(costco_orders_csv),
resolution_rows,
existing_links,
catalog_rows,
)
existing_catalog_rows = read_optional_csv_rows(catalog_csv)
merged_catalog_rows = merge_catalog_rows(
existing_catalog_rows,
[catalog_row_from_canonical(row) for row in canonical_rows],
)
link_rows = apply_manual_resolutions_to_links(link_rows, resolution_rows)
example_rows = build_comparison_examples(purchase_rows)
write_csv_rows(catalog_csv, merged_catalog_rows, CATALOG_FIELDS)
write_csv_rows(links_csv, link_rows, build_canonical_layer.LINK_FIELDS)
write_csv_rows(catalog_csv, catalog_rows, CATALOG_FIELDS)
write_csv_rows(links_csv, link_rows, PRODUCT_LINK_FIELDS)
write_csv_rows(output_csv, purchase_rows, PURCHASE_FIELDS)
write_csv_rows(examples_csv, example_rows, EXAMPLE_FIELDS)
click.echo(
f"wrote {len(purchase_rows)} purchase rows to {output_csv}, "
f"{len(merged_catalog_rows)} catalog rows to {catalog_csv}, "
f"{len(catalog_rows)} catalog rows to {catalog_csv}, "
f"{len(link_rows)} product links to {links_csv}, "
f"and {len(example_rows)} comparison examples to {examples_csv}"
)