# Data Types in Files — Linked to `README.md`

> This file explains the **physical type of every column in every Parquet/CSV file** and links it to the corresponding section in the main documentation `README.md`. Every row contains a direct `README.md:line_number` link for easy navigation.

---

## How to Read the Table

| Layer | Example | Verification Tool |
|---|---|---|
| **Parquet Physical** | `optional binary text (String)` | `pyarrow.parquet.ParquetFile(path).schema` |
| **Arrow** | `text: string` | `pf.schema_arrow` |
| **Hugging Face Features** | `Value(string)` | `pf.metadata.metadata[b'huggingface']` |
| **Pandas dtype** | `string` / `int64` / `float64` | `pd.read_parquet(path).dtypes` |

- `optional` = nullable (`NULL` allowed)
- All strings are `UTF-8`
- All numerics are `little-endian`

---

## 1. `egyptian_arabic` — `Egyptian-Arabic/*.parquet` (23,850,855 rows)

**Linked to:** `README.md:173` (Data Instances §1) + `README.md:397` (Appendix A.1) + `README.md:70` (Dataset Structure)

| Column | Physical Type | Arrow | HF | Pandas | Example |
|---|---|---|---|---|---|
| `text` | `optional binary (String)` | `string` | `Value(string)` | `string` | `عندى كام فكره...` / `November 2011` |

- **Size:** 10 shards, single column → streaming-friendly
- **Usage:** `README.md:110` Language Modeling
- **Verification code:** `README.md:420`

---

## 2. `wikipedia` — `Egyptian_Arabic_Wikipedia_20230101/*.parquet` (728,337)

**Linked to:** `README.md:188` + `README.md:428` (Appendix A.2)

| `text` | `optional binary (String)` | `string` | `Value(string)` | `string` | `علم جنوب السودان ...` (avg 2,292 chars) |

- **Important:** First row contains literal value `text` → filter with `df[df.text!="text"]` (`README.md:525`)
- **No** `huggingface` metadata → inferred automatically

---

## 3. `convs` — `egyptian_arabic_convs/*.parquet` (2,517 conversations)

**Linked to:** `README.md:199` + `README.md:438` (Appendix A.3)

| `text` | `optional group (List) { repeated group list { struct { content: string, role: string } } }` | `list<element: struct<content:string, role:string>>` | `Sequence(Sequence(struct))` | `object` (list of dict) | `[{"role":"user","content":"مين سقراط؟"}, {"role":"model","content":"سقراط كان..."}]` |

- **HF Features:**
```json
"text": [{"content": {"dtype":"string"}, "role": {"dtype":"string"}}]
```
- **Parsing:** `json.loads(example["text"])` → `README.md:211`
- **Usage:** Instruction-tuning / Chatbot `README.md:110`

---

## 4. `speech_whisper` — `egyptian-arabic-speech-dataset/*.parquet` (2,571)

**Linked to:** `README.md:220` + `README.md:457` (Appendix A.4) + `README.md:465` (Whisper recipe)

| Column | Parquet | Arrow | HF | Pandas | Shape |
|---|---|---|---|---|---|
| `input_features` | `optional group List { repeated group List { optional float } }` | `list<element: list<element: float>>` | `Sequence(Sequence(Value(float32)))` | `object` | `(80, 3000)` float32 → `np.stack()` = 0.96MB/row |
| `labels` | `optional group List { repeated group list { optional int64 } }` | `list<element: int64>` | `Sequence(Value(int64))` | `object` | `[50258,50272,50359,50363,...,50257]` |

- **Tokens:** `50258 <sot>`, `50272 <ar>`, `50359 <transcribe>`, `50363 <notimestamps>`, `50257 <eot>` — same as `openai/whisper`
- **Storage:** log-Mel 80 channels × 3000 frames (30s), padding ~ `-0.598` for silence

---

## 5. `english_to_arabic` — `english-to-arabic/*.parquet` (33,299 unique)

**Linked to:** `README.md:237` + `README.md:474` (Appendix A.5) + `README.md:369` (Splits)

| Column | Parquet | Arrow | HF | Pandas | Example |
|---|---|---|---|---|---|
| `Egy` | `optional binary (String)` | `string` | `Value(string)` | `string` | `رضا فين؟` |
| `English` | `optional binary (String)` | `string` | `Value(string)` | `string` | `Where is Rida?` |
| `Egy_Text_Source` | `optional binary (String)` | `string` | `Value(string)` | `string` | `Al-Sabbagh 2023 Mendeley` |

- **Sources share identical type:** ArzEn 13,946 + EGY Songs 6,554 ×2 (duplicate) + NADI 12,799 → `README.md:248`
- **Dedup warning:** Both `Milion_Token*` files are identical → `README.md:525`

---

## 6. `qa_trilingual` — `english-to-arabic/train.csv` (37,149)

**Linked to:** `README.md:262` + `README.md:486` (Appendix A.6) + `dataset_info.yaml:3`

| Column | CSV Raw | Arrow (pandas) | HF Feature | Pandas | Example |
|---|---|---|---|---|---|
| `question` | `string` quoted | `large_string` | `Value(string)` | `string` | `مين اللي اخترع...؟` |
| `answer` | `string` | `large_string` | `Value(string)` | `string` | `تيم بيرنرز لي...` |
| `category` | `string` | `large_string` | `Value(string)` | `string` | `History` (48 values) |
| `sub_category` | `string` | `large_string` | `Value(string)` | `string` | `Discoveries` |
| `language` | `string` enum | `large_string` | `Value(string)` | `string` | `ar`/`ar_eg`/`en` |
| `question_char_length` | `int` | `int64` | `Value(int32)` | `int64` | `48` |
| `answer_char_length` | `int` | `int64` | `Value(int32)` | `int64` | `213` |
| `question_word_count` | `int` | `int64` | `Value(int32)` | `int64` | `8` |
| `answer_word_count` | `int` | `int64` | `Value(int32)` | `int64` | `37` |

- **Distribution:** `ar_eg 12,451 | en 12,432 | ar 12,266` → `README.md:277`
- **Unique:** The only CSV in the corpus

---

## 7-9. `balanced` / `unbalanced` / `uncategorized` — `balanced/*.parquet` etc.

**Linked to:** `README.md:296` + `README.md:504` (Appendix A.7) — **identical 13-column schema**

| Column | Parquet | Arrow | HF | Pandas | Role |
|---|---|---|---|---|---|
| `page_title` | `optional binary (String)` | `string` | `Value(string)` | `string` | title |
| `creation_date` | `optional binary (String)` | `string` | `Value(string)` | `string` | `2020-05-08` **stored as string, not date32** → `pd.to_datetime()` |
| `creator_name` | `optional binary (String)` | `string` | `Value(string)` | `string` | `HitomiAkane` |
| `total_edits` | `optional int64` | `int64` | `Value(int64)` | `int64` | count |
| `total_editors` | `optional int64` | `int64` | `Value(int64)` | `int64` | count |
| `top_editors` | `optional binary (String)` | `string` | `Value(string)` | `string` | **`"['GhalyBot','HitomiAkane']"` as JSON string → needs `ast.literal_eval`** |
| `bots_editors_percentage` | `optional double` | `double` | `Value(float64)` | `float64` | `50.0` |
| `humans_editors_percentage` | `optional double` | `double` | `Value(float64)` | `float64` | `50.0` |
| `total_bytes` | `optional int64` | `int64` | `Value(int64)` | `int64` | bytes |
| `total_chars` | `optional int64` | `int64` | `Value(int64)` | `int64` | chars |
| `total_words` | `optional int64` | `int64` | `Value(int64)` | `int64` | words |
| `page_text` | `optional binary (String)` | `string` | `Value(string)` | `string` | body (461–1002 avg) |
| `label` | `optional binary (String)` | `string` | `Value(string)` | `string` | `Human-generated` / `Template-translated` |

- **Sizes:** balanced 20K (50% / 50%) — unbalanced 166K (93% Template) — uncategorized 569K → `README.md:315`
- **Warning:** `top_editors` looks like list but is **string** → `README.md:526`

---

## 10. `reviews_spam` — `part 1 data/*.parquet` (60,000 — 26 columns)

**Linked to:** `README.md:327` + `README.md:528` (Appendix A.8) + `README.md:477` (Training Recipe D)

| # | Column | Parquet | Arrow | HF | Pandas | Example / Range |
|---|---|---|---|---|---|---|
| 1 | `user_id` | `int64` | `int64` | `Value(int64)` | `int64` | `48592` |
| 2 | `product_id` | `int64` | `int64` | `Value(int64)` | `int64` | `3791` |
| 3 | `original_review` | `binary (String)` | `string` | `Value(string)` | `string` | `Love this place!` (English) |
| 4 | `translated_review` | `binary (String)` | `string` | `Value(string)` | `string` | `بحب المكان ده!` (dialectal) |
| 5 | `normalized_translated_review` | `binary (String)` | `string` | `Value(string)` | `string` | `بحب المكان ده! في وقت من الاوقات` (normalized) |
| 6 | `date` | `binary (String)` | `string` | `Value(string)` | `string` | `2014-03-21` |
| 7 | `rating` | `int64` | `int64` | `Value(int64)` | `int64` | `1-5` |
| 8 | `sentiment_label` | `binary (String)` | `string` | `Value(string)` | `string` | `positive/negative/neutral` |
| 9 | `positive_normalized_score` | `double` | `double` | `Value(float64)` | `float64` | `0.0-1.0` |
| 10 | `neutral_normalized_score` | `double` | `double` | `Value(float64)` | `float64` | `0.0-1.0` |
| 11 | `negative_normalized_score` | `double` | `double` | `Value(float64)` | `float64` | `0.0-1.0` |
| 12 | `spam_hit_score` | `int64` | `int64` | `Value(int64)` | `int64` | `0` (46K) / `1` (2.8K) |
| 13 | `arabic_num_words` | `int64` | `int64` | `Value(int64)` | `int64` | `112` |
| 14 | `entropy1` | `double` | `double` | `Value(float64)` | `float64` | `3.5-8.0` |
| 15 | `entropy2` | `double` | `double` | `Value(float64)` | `float64` | `3.5-8.0` |
| 16 | `first_review_date` | `binary (String)` | `string` | `Value(string)` | `string` | `2012-01-10` |
| 17 | `last_review_date` | `binary (String)` | `string` | `Value(string)` | `string` | `2014-05-01` |
| 18 | `review_gap_days` | `int64` | `int64` | `Value(int64)` | `int64` | `0-2000` |
| 19 | `review_count` | `int64` | `int64` | `Value(int64)` | `int64` | `1-500` |
| 20 | `product_avg_rating` | `double` | `double` | `Value(float64)` | `float64` | `2.5-5.0` |
| 21 | `rating_deviation` | `double` | `double` | `Value(float64)` | `float64` | `-3 .. +3` |
| 22 | `product_first_review_date` | `binary (String)` | `string` | `Value(string)` | `string` | `2011-06-15` |
| 23 | `days_since_first_review` | `int64` | `int64` | `Value(int64)` | `int64` | `0-4000` |
| 24 | `user_tenure_days` | `int64` | `int64` | `Value(int64)` | `int64` | `0-4000` |
| 25 | `label` | `int64` | `int64` | `Value(int64)` | `int64` | `0/1` |
| 26 | `label_str` | `binary (String)` | `string` | `Value(string)` | `string` | `authentic/fake` (50/50) |

- **Note:** `label` (int) and `label_str` (string) are dual encodings of the same binary target → use `label_str` for readability and `label` for loss

---

## Quick Link Matrix (File → Config → README Section)

| File Pattern | Config `load_dataset` | README Section |
|---|---|---|
| `Egyptian-Arabic/*.parquet` | `egyptian_arabic` | `README.md:173` + `README.md:397` |
| `Egyptian_Arabic_Wikipedia_20230101/*.parquet` | `wikipedia` | `README.md:188` + `README.md:428` |
| `egyptian_arabic_convs/*.parquet` | `convs` | `README.md:199` + `README.md:438` |
| `egyptian-arabic-speech-dataset/*.parquet` | `speech_whisper` | `README.md:220` + `README.md:457` |
| `english-to-arabic/*.parquet` | `english_to_arabic` | `README.md:237` + `README.md:474` |
| `english-to-arabic/train.csv` | `qa_trilingual` | `README.md:262` + `README.md:486` |
| `balanced/*.parquet` | `balanced` | `README.md:296` + `README.md:504` |
| `unbalanced/*.parquet` | `unbalanced` | `README.md:296` + `README.md:504` |
| `uncategorized/*.parquet` | `uncategorized` | `README.md:296` + `README.md:504` |
| `part 1 data/*.parquet` | `reviews_spam` | `README.md:327` + `README.md:528` |

---

## Storage & Cross-Link Notes

- All Parquet files are `Snappy`-compressed with dictionary-encoded strings — `README.md:578`
- Files without `b'huggingface'` metadata (e.g., `wikipedia`) are inferred automatically at `load_dataset` time
- To validate a new shard before contributing:
```python
import pyarrow.parquet as pq
pf = pq.ParquetFile("path.parquet")
assert pf.schema_arrow.equals(expected)  # compare with README.md:528
```
- **Widest:** `reviews_spam` (26 columns) → slowest to scan, **Tallest:** `egyptian_arabic` (23M) → use `streaming=True` → `README.md:594`

---

**Summary:** This file is the physical map, and `README.md` is the logical map. Keep them together when uploading to Hugging Face.