Hugging Face's logo

Datasets:
saraNour
/
compliments-reference-db

Languages:
English
Tags:
product-database
food-nutrition
nutri-score
canadian-products
compliments
sensations
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
row_count: struct<input: int64, output: int64, pass: bool>
  child 0, input: int64
  child 1, output: int64
  child 2, pass: bool
original_columns_preserved: struct<missing: list<item: null>, pass: bool>
  child 0, missing: list<item: null>
      child 0, item: null
  child 1, pass: bool
new_columns_added: struct<expected: list<item: string>, actual: list<item: string>, missing: list<item: null>, pass: bo (... 3 chars omitted)
  child 0, expected: list<item: string>
      child 0, item: string
  child 1, actual: list<item: string>
      child 0, item: string
  child 2, missing: list<item: null>
      child 0, item: null
  child 3, pass: bool
brand_norm: struct<unique_values: list<item: string>, pass: bool>
  child 0, unique_values: list<item: string>
      child 0, item: string
  child 1, pass: bool
product_line: struct<unique_values: list<item: string>, pass: bool>
  child 0, unique_values: list<item: string>
      child 0, item: string
  child 1, pass: bool
fat_level: struct<valid_values: list<item: string>, actual_values: list<item: string>, invalid: list<item: null (... 14 chars omitted)
  child 0, valid_values: list<item: string>
      child 0, item: string
  child 1, actual_values: list<item: string>
      child 0, item: string
  child 2, invalid: list<item: null>
      child 0, item: null
  child 3, pass: bool
identity_flags_boolean: struct<pass: bool>
  child 0, pass: bool
core_title: struct<empty_count: int64, pass: string>
  child 0, empty_count: int64
  child 1, pa
...
ass: string>
  child 0, empty_count: int64
  child 1, pass: string
external_id_uniqueness: struct<duplicate_count: int64, pass: string>
  child 0, duplicate_count: int64
  child 1, pass: string
overall: struct<result: string>
  child 0, result: string
version: string
failures: list<item: null>
  child 0, item: null
timestamp: string
result: string
checks: struct<schema: struct<pass: bool, note: string>, external_id_uniqueness: struct<total_rows: int64, n (... 308 chars omitted)
  child 0, schema: struct<pass: bool, note: string>
      child 0, pass: bool
      child 1, note: string
  child 1, external_id_uniqueness: struct<total_rows: int64, null_count: int64, unique_count: int64, duplicate_count: int64, pass: stri (... 3 chars omitted)
      child 0, total_rows: int64
      child 1, null_count: int64
      child 2, unique_count: int64
      child 3, duplicate_count: int64
      child 4, pass: string
  child 2, duplicates: struct<full_row_duplicates: int64, external_id_duplicates: int64, upc_duplicates: int64, title_size_ (... 24 chars omitted)
      child 0, full_row_duplicates: int64
      child 1, external_id_duplicates: int64
      child 2, upc_duplicates: int64
      child 3, title_size_brand_duplicates: int64
  child 3, null_columns: struct<expected_100pct_null: list<item: string>, dropped: list<item: string>>
      child 0, expected_100pct_null: list<item: string>
          child 0, item: string
      child 1, dropped: list<item: string>
          child 0, item: string
to
{'version': Value('string'), 'timestamp': Value('string'), 'result': Value('string'), 'failures': List(Value('null')), 'checks': {'schema': {'pass': Value('bool'), 'note': Value('string')}, 'external_id_uniqueness': {'total_rows': Value('int64'), 'null_count': Value('int64'), 'unique_count': Value('int64'), 'duplicate_count': Value('int64'), 'pass': Value('string')}, 'duplicates': {'full_row_duplicates': Value('int64'), 'external_id_duplicates': Value('int64'), 'upc_duplicates': Value('int64'), 'title_size_brand_duplicates': Value('int64')}, 'null_columns': {'expected_100pct_null': List(Value('string')), 'dropped': List(Value('string'))}}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 489, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              row_count: struct<input: int64, output: int64, pass: bool>
                child 0, input: int64
                child 1, output: int64
                child 2, pass: bool
              original_columns_preserved: struct<missing: list<item: null>, pass: bool>
                child 0, missing: list<item: null>
                    child 0, item: null
                child 1, pass: bool
              new_columns_added: struct<expected: list<item: string>, actual: list<item: string>, missing: list<item: null>, pass: bo (... 3 chars omitted)
                child 0, expected: list<item: string>
                    child 0, item: string
                child 1, actual: list<item: string>
                    child 0, item: string
                child 2, missing: list<item: null>
                    child 0, item: null
                child 3, pass: bool
              brand_norm: struct<unique_values: list<item: string>, pass: bool>
                child 0, unique_values: list<item: string>
                    child 0, item: string
                child 1, pass: bool
              product_line: struct<unique_values: list<item: string>, pass: bool>
                child 0, unique_values: list<item: string>
                    child 0, item: string
                child 1, pass: bool
              fat_level: struct<valid_values: list<item: string>, actual_values: list<item: string>, invalid: list<item: null (... 14 chars omitted)
                child 0, valid_values: list<item: string>
                    child 0, item: string
                child 1, actual_values: list<item: string>
                    child 0, item: string
                child 2, invalid: list<item: null>
                    child 0, item: null
                child 3, pass: bool
              identity_flags_boolean: struct<pass: bool>
                child 0, pass: bool
              core_title: struct<empty_count: int64, pass: string>
                child 0, empty_count: int64
                child 1, pa
              ...
              ass: string>
                child 0, empty_count: int64
                child 1, pass: string
              external_id_uniqueness: struct<duplicate_count: int64, pass: string>
                child 0, duplicate_count: int64
                child 1, pass: string
              overall: struct<result: string>
                child 0, result: string
              version: string
              failures: list<item: null>
                child 0, item: null
              timestamp: string
              result: string
              checks: struct<schema: struct<pass: bool, note: string>, external_id_uniqueness: struct<total_rows: int64, n (... 308 chars omitted)
                child 0, schema: struct<pass: bool, note: string>
                    child 0, pass: bool
                    child 1, note: string
                child 1, external_id_uniqueness: struct<total_rows: int64, null_count: int64, unique_count: int64, duplicate_count: int64, pass: stri (... 3 chars omitted)
                    child 0, total_rows: int64
                    child 1, null_count: int64
                    child 2, unique_count: int64
                    child 3, duplicate_count: int64
                    child 4, pass: string
                child 2, duplicates: struct<full_row_duplicates: int64, external_id_duplicates: int64, upc_duplicates: int64, title_size_ (... 24 chars omitted)
                    child 0, full_row_duplicates: int64
                    child 1, external_id_duplicates: int64
                    child 2, upc_duplicates: int64
                    child 3, title_size_brand_duplicates: int64
                child 3, null_columns: struct<expected_100pct_null: list<item: string>, dropped: list<item: string>>
                    child 0, expected_100pct_null: list<item: string>
                        child 0, item: string
                    child 1, dropped: list<item: string>
                        child 0, item: string
              to
              {'version': Value('string'), 'timestamp': Value('string'), 'result': Value('string'), 'failures': List(Value('null')), 'checks': {'schema': {'pass': Value('bool'), 'note': Value('string')}, 'external_id_uniqueness': {'total_rows': Value('int64'), 'null_count': Value('int64'), 'unique_count': Value('int64'), 'duplicate_count': Value('int64'), 'pass': Value('string')}, 'duplicates': {'full_row_duplicates': Value('int64'), 'external_id_duplicates': Value('int64'), 'upc_duplicates': Value('int64'), 'title_size_brand_duplicates': Value('int64')}, 'null_columns': {'expected_100pct_null': List(Value('string')), 'dropped': List(Value('string'))}}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Compliments Reference DB

A clean, reproducible product reference database built from a single authoritative dataset for Sobeys Inc.'s Compliments and Sensations private-label brands.

Pipeline: 6 deterministic phases, no LLM, full traceability.

Authoritative Source: saraNour/compliments-brand

Published DB: saraNour/compliments-reference-db (Public)

Quick Start 

from huggingface_hub import hf_hub_download
import pandas as pd

# Load any phase output
path = hf_hub_download("saraNour/compliments-reference-db", "phase6/outputs/phase6_product_scores.parquet", repo_type="dataset")
df = pd.read_parquet(path)

Pipeline Overview

Phase Name Input Output Key Metric
1 Data Quality Foundation products.parquet (4,440 × 16) phase1_output.parquet (4,440 × 14) 100% null cols dropped
2 Semantic Normalization phase1_output (4,440 × 14) phase2_output.parquet (4,440 × 34) 20 new columns
3 Product Domain + Taxonomy + Grouping phase2_output (4,440 × 34) reference_product_catalog.csv (3,691 groups) 19 taxonomy categories
4 Variant Assignment phase3 mapping + phase2 product_variant_mapping.parquet (4,440 × 11) 4,299 variants
5 Nutrition Integration nutrition.parquet + phase3 + phase4 nutrition_per_100g.parquet 100% match rate
6 Nutri-Score + Agribalyse phase5 outputs phase6_product_scores.parquet (4,440) 364 scored (8.2%)

Data Sources

Authoritative Inputs

File Rows × Cols Source
products.parquet 4,440 × 16 saraNour/compliments-brand
nutrition.parquet 4,440 × 21 saraNour/compliments-brand

External References

Resource URL Purpose
Google Product Category taxonomy https://www.google.com/basepages/producttype/taxonomy.en-US.txt GPC mapping
Nutri-Score 2023 algorithm https://www.eurofins.de/food-analysis/other-services/nutri-score/ Nutri-Score calculation
Santé Publique France FAQ v21.Dec.2023 (via Eurofins) Algorithm verification
Agribalyse v3.2 CIQUAL database Environmental category mapping

Phase 1 — Data Quality Foundation

Goal: Download authoritative data, validate schema, clean, and produce a reproducible foundation.

What It Does

  1. Downloads products.parquet from HuggingFace
  2. Validates schema (16 expected columns, 4,440 rows)
  3. Audits raw data (nulls, duplicates, UPC patterns, brands)
  4. Cleans: whitespace normalization, empty→null, brand/title cleaning
  5. Drops 100% null columns (size_per_unit, size_total)
  6. Produces clean output + validation + statistics

Cleaning Operations

  • String normalization: trim whitespace, collapse repeated spaces, empty strings → null
  • Brand cleaning: whitespace trim, ALL CAPS → Title Case (preserve raw)
  • UPC validation: format check, null count, duplicate analysis, reused UPC detection
  • Title cleaning: whitespace normalization
  • Null column handling: documented removal of 100% null columns

Output: phase1_output.parquet (4,440 × 14)

Column Type Description
upc string Product UPC (3,271 unique, 774 reused, 1 null)
external_id string Product-level identifier (4,440 unique)
brand string Raw brand name
brand_clean string Cleaned brand name
title string Raw product title
title_clean string Cleaned title
price float Product price
price_currency string Currency code
size string Raw size string
size_amount float Numeric size amount
size_unit string Size unit (g, mL, etc.)
size_qty int Quantity
size_unit_norm string Normalized size unit
image_url string Product image URL
source string Data source
source_url string Source URL

Validation

  • external_id uniqueness: PASS
  • No full-row duplicates: PASS
  • Schema validation: PASS

Phase 2 — Semantic Normalization

Goal: Normalize brands, extract identity attributes from original titles, parse variants, build core titles and identity hashes.

What It Does

  1. Normalizes 9 raw brand variants → 2 brands (Compliments, Sensations) + 7 product lines
  2. Extracts 9 boolean identity flags from ORIGINAL title (before normalization)
  3. Extracts fat level and fat percentage
  4. Extracts flavour and formulation keywords
  5. Parses variant attributes from size strings
  6. Extracts core titles (strip brand prefix + size)
  7. Builds deterministic identity hash

Brand Normalization

Raw Brand Normalized Product Line
Compliments Organic Compliments Organic
Compliments Balance Compliments Balance
Compliments Naturally Simple Compliments Naturally Simple
Compliments Green Care Compliments Green
Compliments Little Ones Compliments Little Ones
Sensations Sensations Sensations
Compliments Compliments Core

Identity Flags Extracted

  • is_organic — Organic products
  • is_gluten_free — Gluten-free products
  • is_naturally_simple — Naturally Simple line
  • is_sugar_free — Sugar-free / no sugar added / unsweetened
  • is_unsalted — Unsalted / no salt
  • is_lactose_free — Lactose-free
  • is_peanut_free — Peanut-free
  • is_plant_based — Plant-based
  • is_reduced_sodium — Reduced sodium / low sodium

Output: phase2_output.parquet (4,440 × 34)

All Phase 1 columns PLUS:

Column Type Description
brand_norm string Normalized brand name
product_line string Product line (Core, Organic, Balance, etc.)
is_organic bool Organic flag
is_gluten_free bool Gluten-free flag
is_naturally_simple bool Naturally Simple flag
is_sugar_free bool Sugar-free flag
is_unsalted bool Unsalted flag
is_lactose_free bool Lactose-free flag
is_peanut_free bool Peanut-free flag
is_plant_based bool Plant-based flag
is_reduced_sodium bool Reduced sodium flag
fat_level string regular / reduced_fat / fat_free
fat_percentage float Numeric fat percentage (if in title)
flavour list Extracted flavour keywords
formulation list Extracted formulation keywords
variant_attributes JSON Parsed size/package attributes
core_title string Title stripped of brand prefix and size
identity_hash string Deterministic hash of all identity attributes

Phase 3 — Product Domain + Taxonomy + Product Grouping

Goal: Classify food/non-food, assign taxonomy, group products by identity, generate reference catalog.

What It Does

  1. Classifies products as food / non_food / unknown using keyword rules
  2. Assigns taxonomy category (19 categories) using keyword rules
  3. Groups products by identity hash + domain + normalized core title
  4. Detects ambiguous cases (multiple flavours, formulations, etc. in same group)
  5. Generates reference product catalog (one row per group)
  6. Generates product group mapping (one row per product)
  7. Maps internal taxonomy to Google Product Category (GPC)
  8. Builds separate non-food products table

Food/Non-Food Classification Results

Domain Count %
food 2,814 63.4%
unknown 1,232 27.7%
non_food 394 8.9%

Taxonomy Categories (19)

Category Description GPC Mapping
DAIRY Yogurt, milk, cheese, butter Dairy Products (428)
MEAT_SEAFOOD Chicken, beef, fish, seafood Meat, Seafood & Eggs (432)
BEVERAGES Juice, coffee, tea, drinks Beverages (413)
BAKERY Bread, bagel, muffin, cookie Bakery (1876)
FROZEN Frozen food, ice cream, pizza Frozen Desserts (5788)
PRODUCE Fruits, vegetables, olives Fruits & Vegetables (430)
CONDIMENTS_SAUCES Sauce, ketchup, mustard, oil Condiments & Sauces (427)
CONFECTIONERY Chocolate, candy, gummy Candy & Chocolate (4748)
SNACKS Chip, cracker, popcorn, nut Snack Foods (423)
PASTA_RICE Pasta, noodle, rice Grains, Rice & Cereal (431)
BREAKFAST Cereal, oat, granola Cereal & Granola (4689)
CANNED_GOODS Canned products Cooking & Baking Ingredients (2660)
GENERAL_GROCERY Default catch-all Food Items (422)
HOUSEHOLD_CLEANING Detergent, laundry Cleaning Supplies
HEALTH_REMEDIES Medication, vitamins Health Care
PERSONAL_CARE Shampoo, deodorant Personal Care
BABY_CARE Baby products Baby & Toddler Food
PET_FOOD Cat/dog food Pet Food
HOUSEHOLD_SUPPLIES Kitchenware, bulbs Home Décor

Product Grouping

  • Group key: identity_hash | product_domain | normalized_core_title
  • Group ID: Deterministic UUID5 from group key
  • Groups: 3,691 unique groups
  • Ambiguous cases: 303 (multiple core titles, flavours, or formulations in same group)

Output Tables

reference_product_catalog.csv (3,691 rows)

Column Description
group_id Deterministic UUID5
group_name Most frequent core_title in group
brand Brand name
product_domain food / non_food / unknown
reference_db_taxonomy Category classification
identity_hash Identity hash from Phase 2
product_count Number of products in group
unique_upcs Number of unique UPCs

product_group_mapping.csv (4,440 rows)

Column Description
upc Product UPC
external_id External identifier
group_id Group UUID
group_name Group name
core_title Normalized core title
original_title Original product title
brand Brand name
size Product size
variant_attributes Variant attributes JSON
identity_hash Identity hash
product_domain food / non_food / unknown
reference_db_taxonomy Category
product_line Product line
is_organic ... is_reduced_sodium Identity flags
fat_level Fat level
fat_percentage Fat percentage
flavour Flavour keywords
formulation Formulation keywords
source Data source
source_url Source URL

taxonomy_gpc_mapping.csv (19 rows)

Column Description
reference_db_taxonomy Internal taxonomy category
gpc_id Google Product Category ID
gpc_name GPC category name
gpc_full_path Full GPC hierarchy path

non_food_products.csv (394 rows)

Column Description
external_id External identifier
upc Product UPC
group_id Group UUID
group_name Group name
core_title Normalized core title
original_title Original product title
brand Brand name
size Product size
reference_db_taxonomy Category
product_line Product line
source Data source
source_url Source URL

Phase 4 — Variant Assignment

Goal: Assign variant IDs to products within each group based on size/package attributes.

What It Does

  1. Loads Phase 3 mapping + Phase 2 variant data
  2. Builds variant key from size/package attributes
  3. Generates deterministic variant ID (UUID5 from group_id + variant_key)
  4. Builds variant table (one row per unique variant)
  5. Builds product → variant mapping

Variant Key Structure 

amt{amount}|unit{unit}|qty{qty}|mult{multiplier}

Examples:

  • amt500.0|unitg — 500g product
  • amt1.0|unitkg — 1kg product
  • amt500.0|unitg|qty2 — 2 × 500g multipack
  • count20 — 20-count pack
  • nosize — No size information

Output Tables

product_variant_mapping.parquet (4,440 × 11)

Column Description
external_id External identifier
upc Product UPC
group_id Group UUID
variant_id Variant UUID
core_title Normalized core title
original_title Original product title
brand Brand name
size Product size
variant_key Variant key string
source Data source
source_url Source URL

reference_product_variants.parquet (4,299 rows)

Column Description
variant_id Variant UUID
group_id Group UUID
group_name Group name
brand Brand name
product_line Product line
product_domain food / non_food / unknown
reference_db_taxonomy Category
core_title Normalized core title
variant_key Variant key string
size Size string
size_amount Numeric size
size_unit Size unit
flavour Flavour keywords
formulation Formulation keywords
fat_level Fat level
fat_percentage Fat percentage
product_count Products in this variant
unique_upcs Unique UPCs in this variant

Phase 5 — Nutrition Integration

Goal: Match nutrition data to products, normalize per-100g values, validate 1:1 relationship.

What It Does

  1. Loads nutrition data (4,440 × 21) from HuggingFace
  2. Cleans and validates nutrition (flags MISSING, SUSPICIOUS)
  3. Normalizes nutrition to per-100g/mL values
  4. Matches products to nutrition by external_id (100% deterministic)
  5. Joins with Phase 3 (group_id) and Phase 4 (variant_id)

Normalization Methods

Method Count Description
direct_100g 150 Serving size is 100g/mL
scaled_to_100g 1,351 Scaled from serving size to 100g/mL
not_normalized 2,939 Ambiguous unit or missing serving size

Nutrition Quality Status

Status Count Description
VALID 4,076 All nutrition fields present
MISSING 364 All nutrition fields null (non-food)
SUSPICIOUS 0 Values exceed thresholds

Output Tables

nutrition_per_100g.parquet (4,440 rows)

Column Description
external_id External identifier
upc Product UPC
serving_size Raw serving size string
serving_amount Parsed numeric amount
serving_unit Parsed unit (g, mL, etc.)
normalization_method direct_100g / scaled_to_100g / not_normalized
normalization_reason Explanation of normalization
calories_per_100g Calories per 100g
carbohydrate_g_per_100g Carbohydrates per 100g
sugars_g_per_100g Sugars per 100g
sodium_mg_per_100g Sodium per 100g
fat_g_per_100g Fat per 100g
saturated_fat_g_per_100g Saturated fat per 100g
protein_g_per_100g Protein per 100g
fibre_g_per_100g Fibre per 100g
... (17 nutrition fields total)

product_nutrition_mapping.parquet (4,440 rows)

Column Description
external_id External identifier
upc Product UPC
group_id Group UUID
variant_id Variant UUID
title Product title
serving_size Raw serving size
calories ... cholesterol_mg 17 nutrition fields
match_method external_id_exact
match_confidence HIGH
nutrition_quality_status VALID / MISSING / SUSPICIOUS
nutrition_quality_flags Quality flags

Phase 6 — Nutri-Score 2023 + Agribalyse Mapping

Goal: Calculate Nutri-Score 2023 for eligible food products, map to Agribalyse categories.

Algorithm

Nutri-Score 2023 (updated algorithm)

Eligibility Criteria

A product is eligible if ALL of:

  1. product_domain = "food"
  2. nutrition_quality_status = "VALID"
  3. ALL 6 required per-100g fields present:
    • calories_per_100g
    • sugars_g_per_100g
    • saturated_fat_g_per_100g
    • sodium_mg_per_100g
    • fibre_g_per_100g
    • protein_g_per_100g

Results

Metric Value
Total products 4,440
Eligible 364 (8.2%)
Not eligible 4,076 (91.8%)

Nutri-Score Grade Distribution (among eligible)

Grade Score Range Count % of Eligible
A ≤ 0 31 8.5%
B 0–2 42 11.5%
C 2–10 135 37.1%
D 10–18 83 22.8%
E > 18 73 20.1%

Exclusion Reasons

Reason Count
NON_FOOD 394
UNKNOWN_DOMAIN 1,232
MISSING_NUTRITION 364
MISSING_REQUIRED_FIELDS 2,086

Agribalyse Category Mapping

Confidence Count Description
HIGH 1,908 DAIRY, MEAT_SEAFOOD, PRODUCE, BEVERAGES, CONFECTIONERY
MEDIUM 885 BAKERY, BREAKFAST, SNACKS, CONDIMENTS, FROZEN, PASTA_RICE
LOW 1,310 GENERAL_GROCERY
NONE 337 Non-food, unknown, health, household, personal care

FVL (Fruits, Vegetables, Legumes) Estimation

FVL % Count Source
90% ~200 PRODUCE category estimate
0% ~4,240 All other categories

Output Tables

phase6_product_scores.parquet (4,440 rows)

Column Description
external_id External identifier
group_id Group UUID
variant_id Variant UUID
upc Product UPC
core_title Normalized core title
product_domain food / non_food / unknown
reference_db_taxonomy Category
nutrition_quality_status VALID / MISSING / SUSPICIOUS
score_eligibility ELIGIBLE / NOT_ELIGIBLE
score_exclusion_reason Reason for exclusion
calories_per_100g Calories per 100g
energy_kj_100g Energy in kJ per 100g
sugars_g_per_100g Sugars per 100g
saturated_fat_g_per_100g Saturated fat per 100g
sodium_mg_per_100g Sodium per 100g
salt_g_100g Salt per 100g
fibre_g_per_100g Fibre per 100g
protein_g_per_100g Protein per 100g
fvl_percent FVL percentage
negative_points Nutri-Score negative points
positive_points Nutri-Score positive points
nutri_score_raw Raw Nutri-Score (-10 to 40)
nutri_score_grade A / B / C / D / E
nutri_score_calculated boolean

phase6_agribalyse_mapping.parquet (4,440 rows)

Column Description
external_id External identifier
group_id Group UUID
reference_db_taxonomy Category
product_domain food / non_food / unknown
agribalyse_category Agribalyse category name
agribalyse_ciqual CIQUAL code pattern
mapping_confidence HIGH / MEDIUM / LOW / NONE

phase6_score_exclusions.parquet (4,076 rows)

Column Description
external_id External identifier
group_id Group UUID
product_domain food / non_food / unknown
reference_db_taxonomy Category
nutrition_quality_status MISSING / SUSPICIOUS
score_exclusion_reason Detailed exclusion reason

Repository Structure 

compliments-reference-db/
├── phase1/
│   ├── src/phase1.py
│   ├── outputs/phase1_output.parquet
│   ├── notebooks/phase1.ipynb
│   ├── validation/
│   └── statistics/
├── phase2/
│   ├── src/phase2.py
│   ├── outputs/phase2_output.parquet
│   ├── notebooks/phase2.ipynb
│   ├── validation/
│   └── statistics/
├── phase3/
│   ├── src/phase3.py
│   ├── outputs/
│   │   ├── reference_product_catalog.csv
│   │   ├── product_group_mapping.csv
│   │   ├── taxonomy_gpc_mapping.csv
│   │   ├── non_food_products.csv
│   │   ├── ambiguous_cases.csv
│   │   └── unknown_products.csv
│   ├── notebooks/phase3.ipynb
│   ├── validation/
│   └── statistics/
├── phase4/
│   ├── src/phase4.py
│   ├── outputs/
│   │   ├── product_variant_mapping.parquet
│   │   ├── reference_product_variants.parquet
│   │   └── reference_product_variant_summary.parquet
│   ├── notebooks/phase4.ipynb
│   ├── validation/
│   └── statistics/
├── phase5/
│   ├── src/phase5.py
│   ├── outputs/
│   │   ├── nutrition_per_100g.parquet
│   │   ├── nutrition_cleaned.parquet
│   │   └── product_nutrition_mapping.parquet
│   ├── notebooks/phase5.ipynb
│   ├── validation/
│   └── statistics/
├── phase6/
│   ├── src/phase6.py
│   ├── outputs/
│   │   ├── phase6_product_scores.parquet
│   │   ├── phase6_agribalyse_mapping.parquet
│   │   ├── phase6_score_exclusions.parquet
│   │   ├── phase6_review_queue.parquet
│   │   └── phase6_summary.parquet
│   ├── notebooks/phase6.ipynb
│   └── validation/
├── code/
│   ├── src/           ← All 6 scripts consolidated
│   └── notebooks/     ← All 6 notebooks with Input/Output/Goal headers
├── shared/
└── archive/

Tools & Technologies

Tool Purpose
Python 3.10+ Core language
pandas Data manipulation
numpy Numerical operations
huggingface_hub Dataset loading & publishing
UUID5 (uuid module) Deterministic ID generation
regex (re module) Keyword-based classification
JSON Structured metadata

External Websites Used

Website Purpose
https://huggingface.co/datasets/saraNour/compliments-brand Authoritative source data
https://huggingface.co/datasets/saraNour/compliments-reference-db Published reference DB
https://www.google.com/basepages/producttype/taxonomy.en-US.txt Google Product Category taxonomy
https://www.eurofins.de/food-analysis/other-services/nutri-score/ Nutri-Score 2023 algorithm

Design Principles

  1. Single authoritative source — Only products.parquet and nutrition.parquet from HuggingFace
  2. No LLM — All 6 phases are 100% deterministic/rule-based
  3. Full traceability — Every transformation documented, raw values preserved
  4. Reproducibility — Deterministic UUIDs, deterministic rules, versioned outputs
  5. Validation at every phase — Schema checks, duplicate checks, regression tests
  6. Clean separation — Each phase has clear input/output boundary

Key Numbers Summary

Metric Value
Total products 4,440
Unique external_ids 4,440
Unique UPCs 3,271
Reused UPC values 774
Product groups 3,691
Variants 4,299
Food products 2,814
Non-food products 394
Unknown domain 1,232
Taxonomy categories 19
Nutrition match rate 100%
Nutri-Score eligible 364 (8.2%)
Nutri-Score A 31
Nutri-Score B 42
Nutri-Score C 135
Nutri-Score D 83
Nutri-Score E 73

License

Internal research use. Data sourced from Sobeys Inc. Compliments/Sensations product lines.

Version History

Date Phase Version Change Impact
2026-08-01 Phase 2 2.0.0 → 2.1.0 Fixed extract_flavours substring matching bug lime: 4,440 → 32, apple: 111 → 98
2026-08-01 Phase 3 1.2.0 reference_db_taxonomy rename + GPC mapping 19 taxonomy categories mapped to GPC
2026-07-31 Phase 4 1.0.0 Variant assignment 4,299 variants
2026-07-31 Phase 5 1.0.0 Nutrition integration 100% match rate
2026-07-31 Phase 6 1.0.0 Nutri-Score 2023 + Agribalyse 364 scored

Phase 2 v2.1.0 — Flavour Extraction Bug Fix (2026-08-01)

Root cause: The brand name "Compliments" contains the substring "lime" (comp**lime**nts). The extract_flavours function used substring matching (kw in t), causing every product in the database to receive ['lime'] as a flavour. A secondary issue: "apple" matched inside "pineapple".

Fix: Changed from substring matching to word-boundary regex (\bkeyword\b) with plural handling for irregular plurals (berry/berries, peach/peaches, mango/mangos).

Impact on flavour extraction:

Metric Before (v2.0.0) After (v2.1.0)
Products with flavour 4,440 (100%) 868 (19.5%)
Products without flavour 0 3,572 (80.5%)
lime count 4,440 (all false) 32 (all genuine)
apple count ~111 (13 false) 98 (0 false)
False positives ~4,440 0

Downstream impact: All downstream phases (3–6) were re-run with the corrected Phase 2 output. Group count (3,691), variant count (4,299), and all other structural metrics remained unchanged. All validation checks and 12 regression tests pass.
Downloads last month
217
Total file size:
37.3 MB