adriens Claude Sonnet 4.6 commited on
Commit ·
7dec4dd
1
Parent(s): 7d0f4c2
Initial release — mobitag audit SMS dataset
Browse filesSplit temporel (train < 2025-12-01, test >= 2025-12-01), compression ZSTD,
types corrigés (year/month/hour en int). ~2.46M lignes, 2 splits Parquet.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- .gitignore +3 -0
- README.md +104 -0
- data/test-00000-of-00001.parquet +3 -0
- data/train-00000-of-00001.parquet +3 -0
- prepare_dataset.py +75 -0
.gitignore
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
mobitag_audit.parquet
|
| 2 |
+
__pycache__/
|
| 3 |
+
*.py[cod]
|
README.md
ADDED
|
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
task_categories:
|
| 4 |
+
- tabular-regression
|
| 5 |
+
- time-series-forecasting
|
| 6 |
+
language:
|
| 7 |
+
- fr
|
| 8 |
+
tags:
|
| 9 |
+
- telecom
|
| 10 |
+
- sms
|
| 11 |
+
- new-caledonia
|
| 12 |
+
- mobitag
|
| 13 |
+
- time-series
|
| 14 |
+
- audit
|
| 15 |
+
pretty_name: "Mobitag Audit — SMS Activity New Caledonia"
|
| 16 |
+
size_categories:
|
| 17 |
+
- 1M<n<10M
|
| 18 |
+
configs:
|
| 19 |
+
- config_name: default
|
| 20 |
+
data_files:
|
| 21 |
+
- split: train
|
| 22 |
+
path: data/train-*.parquet
|
| 23 |
+
- split: test
|
| 24 |
+
path: data/test-*.parquet
|
| 25 |
+
---
|
| 26 |
+
|
| 27 |
+
# Mobitag Audit — SMS Activity New Caledonia
|
| 28 |
+
|
| 29 |
+
Dataset d'audit d'activité SMS collecté en Nouvelle-Calédonie, couvrant la période **avril 2023 – juin 2026** (~2.46M événements).
|
| 30 |
+
|
| 31 |
+
## Contenu
|
| 32 |
+
|
| 33 |
+
| Colonne | Type | Description |
|
| 34 |
+
|---|---|---|
|
| 35 |
+
| `date` | timestamp | Horodatage de l'événement SMS |
|
| 36 |
+
| `message_length` | int32 | Longueur du message en caractères |
|
| 37 |
+
| `day_name` | string | Jour de la semaine (Monday…Sunday) |
|
| 38 |
+
| `year` | int16 | Année |
|
| 39 |
+
| `year_month` | string | Année-mois (ex : `2024-07`) |
|
| 40 |
+
| `month` | int8 | Mois (1–12) |
|
| 41 |
+
| `hour` | int8 | Heure de la journée (0–23) |
|
| 42 |
+
|
| 43 |
+
## Splits
|
| 44 |
+
|
| 45 |
+
| Split | Période | Lignes |
|
| 46 |
+
|---|---|---|
|
| 47 |
+
| `train` | 2023-04 → 2025-12 | ~2 191 000 |
|
| 48 |
+
| `test` | 2026-01 → 2026-06 | ~270 000 |
|
| 49 |
+
|
| 50 |
+
Le split est **temporel** : les données de test sont postérieures aux données d'entraînement, ce qui reflète un usage réaliste de prévision.
|
| 51 |
+
|
| 52 |
+
## Tâches ML suggérées
|
| 53 |
+
|
| 54 |
+
### 1. Régression — prédire `message_length`
|
| 55 |
+
Utiliser les features temporelles (`hour`, `day_name`, `month`, `year`) pour prédire la longueur d'un SMS.
|
| 56 |
+
|
| 57 |
+
```python
|
| 58 |
+
from datasets import load_dataset
|
| 59 |
+
|
| 60 |
+
ds = load_dataset("YOUR_USERNAME/mobitag-hf")
|
| 61 |
+
train_df = ds["train"].to_pandas()
|
| 62 |
+
|
| 63 |
+
X = train_df[["hour", "month", "year", "day_name"]]
|
| 64 |
+
y = train_df["message_length"]
|
| 65 |
+
```
|
| 66 |
+
|
| 67 |
+
### 2. Prévision de série temporelle — pics d'activité
|
| 68 |
+
Agréger par heure et prédire le volume d'activité SMS.
|
| 69 |
+
|
| 70 |
+
```python
|
| 71 |
+
import pandas as pd
|
| 72 |
+
from datasets import load_dataset
|
| 73 |
+
|
| 74 |
+
ds = load_dataset("YOUR_USERNAME/mobitag-hf")
|
| 75 |
+
df = ds["train"].to_pandas()
|
| 76 |
+
|
| 77 |
+
hourly = (
|
| 78 |
+
df.set_index("date")
|
| 79 |
+
.resample("h")["message_length"]
|
| 80 |
+
.agg(["count", "mean", "sum"])
|
| 81 |
+
.reset_index()
|
| 82 |
+
)
|
| 83 |
+
```
|
| 84 |
+
|
| 85 |
+
## Utilisation rapide
|
| 86 |
+
|
| 87 |
+
```python
|
| 88 |
+
from datasets import load_dataset
|
| 89 |
+
|
| 90 |
+
ds = load_dataset("YOUR_USERNAME/mobitag-hf")
|
| 91 |
+
print(ds)
|
| 92 |
+
# DatasetDict({
|
| 93 |
+
# train: Dataset({features: [...], num_rows: ~2191000}),
|
| 94 |
+
# test: Dataset({features: [...], num_rows: ~270000})
|
| 95 |
+
# })
|
| 96 |
+
```
|
| 97 |
+
|
| 98 |
+
## Contexte
|
| 99 |
+
|
| 100 |
+
Ce dataset est produit dans le cadre du projet **Mobitag**, initiative de valorisation des données télécom en Nouvelle-Calédonie.
|
| 101 |
+
|
| 102 |
+
## Licence
|
| 103 |
+
|
| 104 |
+
Les données sont soumises aux conditions d'utilisation du projet Mobitag. Contact : voir le profil de l'organisation sur Hugging Face.
|
data/test-00000-of-00001.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e971804935e4ae8e7fc5ce16c26ef0cb5600ae82a17801f7be6712f7a5d6131f
|
| 3 |
+
size 2153497
|
data/train-00000-of-00001.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ba69667b5b0ef9973fb80462ba49aac9eba354c93fe8d8be59dbaa5308165043
|
| 3 |
+
size 13761016
|
prepare_dataset.py
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
Prépare le dataset mobitag_audit pour publication sur Hugging Face.
|
| 4 |
+
|
| 5 |
+
- Corrige les types de colonnes
|
| 6 |
+
- Split temporel dynamique : test = les TEST_MONTHS derniers mois complets
|
| 7 |
+
- Compression ZSTD niveau 3
|
| 8 |
+
- Écrit dans data/train-*.parquet et data/test-*.parquet
|
| 9 |
+
|
| 10 |
+
Usage : remplacer mobitag_audit.parquet et relancer le script.
|
| 11 |
+
"""
|
| 12 |
+
|
| 13 |
+
import pathlib
|
| 14 |
+
import pandas as pd
|
| 15 |
+
import pyarrow as pa
|
| 16 |
+
import pyarrow.parquet as pq
|
| 17 |
+
from dateutil.relativedelta import relativedelta
|
| 18 |
+
|
| 19 |
+
TEST_MONTHS = 6 # nombre de mois complets gardés pour le test
|
| 20 |
+
INPUT_FILE = "mobitag_audit.parquet"
|
| 21 |
+
OUTPUT_DIR = pathlib.Path("data")
|
| 22 |
+
|
| 23 |
+
WRITE_OPTIONS = dict(
|
| 24 |
+
compression="zstd",
|
| 25 |
+
compression_level=3,
|
| 26 |
+
use_dictionary=True,
|
| 27 |
+
write_statistics=True,
|
| 28 |
+
row_group_size=100_000,
|
| 29 |
+
)
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
def fix_types(df: pd.DataFrame) -> pd.DataFrame:
|
| 33 |
+
df["year"] = df["year"].astype("int16")
|
| 34 |
+
df["month"] = df["month"].astype("int8")
|
| 35 |
+
df["hour"] = df["hour"].astype("int8")
|
| 36 |
+
return df
|
| 37 |
+
|
| 38 |
+
|
| 39 |
+
def write_split(df: pd.DataFrame, path: pathlib.Path) -> None:
|
| 40 |
+
table = pa.Table.from_pandas(df, preserve_index=False)
|
| 41 |
+
pq.write_table(table, path, **WRITE_OPTIONS)
|
| 42 |
+
size_mb = path.stat().st_size / 1024 / 1024
|
| 43 |
+
print(f" {path.name}: {len(df):,} lignes — {size_mb:.2f} MB")
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
def main() -> None:
|
| 47 |
+
print(f"Lecture de {INPUT_FILE}...")
|
| 48 |
+
df = pd.read_parquet(INPUT_FILE)
|
| 49 |
+
df = fix_types(df)
|
| 50 |
+
|
| 51 |
+
# Début du 1er jour du mois courant (incomplet) → on remonte TEST_MONTHS en arrière
|
| 52 |
+
latest = df["date"].max()
|
| 53 |
+
current_month_start = latest.replace(day=1, hour=0, minute=0, second=0, microsecond=0)
|
| 54 |
+
split_date = current_month_start - relativedelta(months=TEST_MONTHS)
|
| 55 |
+
print(f"Mois le plus récent : {latest.strftime('%Y-%m')}")
|
| 56 |
+
print(f"Split dynamique : train < {split_date.date()} | test >= {split_date.date()}")
|
| 57 |
+
|
| 58 |
+
train = df[df["date"] < split_date].copy()
|
| 59 |
+
test = df[df["date"] >= split_date].copy()
|
| 60 |
+
|
| 61 |
+
print(f" train : {len(train):,} lignes ({len(train)/len(df)*100:.1f}%)")
|
| 62 |
+
print(f" test : {len(test):,} lignes ({len(test)/len(df)*100:.1f}%)")
|
| 63 |
+
|
| 64 |
+
OUTPUT_DIR.mkdir(exist_ok=True)
|
| 65 |
+
|
| 66 |
+
print("\nÉcriture des fichiers...")
|
| 67 |
+
write_split(train, OUTPUT_DIR / "train-00000-of-00001.parquet")
|
| 68 |
+
write_split(test, OUTPUT_DIR / "test-00000-of-00001.parquet")
|
| 69 |
+
|
| 70 |
+
print("\nDone. Structure prête pour `huggingface-cli upload`.")
|
| 71 |
+
print(" → huggingface-cli upload TON_USERNAME/mobitag-hf . --repo-type dataset")
|
| 72 |
+
|
| 73 |
+
|
| 74 |
+
if __name__ == "__main__":
|
| 75 |
+
main()
|