Datasets:

Phazel
/

tehran-rent-asks

---
license: cc-by-sa-4.0
language:
- fa
- en
pretty_name: Tehran & Karaj apartment rent asks (Divar)
size_categories:
- 10K<n<100K
task_categories:
- tabular-regression
tags:
- real-estate
- rent
- tehran
- iran
- tabular
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train.parquet
  - split: test
    path: data/test.parquet
---
# Tehran & Karaj apartment rent asks (Divar)

Asking rents for whole flats on yearly leases, listed on [Divar](https://divar.ir) in 493 neighborhoods of Tehran
and Karaj. The Tehran part covers all 22 municipal districts, split into five zones. The data comes from the weekly
crawls behind [rent.nlogn.ir](https://rent.nlogn.ir), snapshots 2026-10-09, 2026-10-13, 2026-10-15, 2026-10-19, 2026-10-20, 2026-10-22, 2026-10-27.

These are asking prices from the ads. Signed leases usually close somewhat lower. All money is in million toman.

## Task

The target is `log_ppm`, the natural log of the price per m² after converting the ad to a full deposit (رهن کامل):
`ppm = (deposit_m + rent_m / 0.03) / area_m2`. The 3% monthly rate is the median that Divar
posters set on the convertible-rent slider. `equiv_m` is the same price for the whole flat.

`deposit_m`, `rent_m`, `equiv_m` and `ppm` restate the target in other units, so leave them out of the inputs.

## Splits

The split is forward in time by publish date. `test` holds the ads published in the last 7 days,
after 2026-10-20 up to 2026-10-27, and `train` holds everything published on or before 2026-10-20.

| Split | Rows |
|---|---|
| train | 33,501 |
| test | 9,432 |

Agencies often repost the same flat. For cross-validation on `train`, group folds by `group` (neighborhood, build year
and floor) so the copies stay in one fold.

## Columns

| Column | Type | Meaning |
|---|---|---|
| token, url | str | Divar ad id and link |
| city, zone | str | `tehran`/`karaj`; site zone (west, east, north, center, south, karaj) |
| hood, hood_name | str | Divar neighborhood slug and Persian name |
| hood_lat, hood_lon | float | Divar's public centroid of the neighborhood (not the ad's pin) |
| area_m2 | int | floor area |
| year_built | int | Jalali build year; Divar caps the oldest at 1370 ("1370 or before") |
| age | int | Jalali year at `last_seen` − `year_built` |
| floor, floors_total, rooms | int | floor (0 = ground; may be null), building floors (often null), bedrooms |
| elevator, parking, storage | bool | Divar's structured amenities |
| renovated, complex | bool | ad text mentions renovation / a complex (مجتمع، شهرک، برج …) |
| convertible | bool | the poster allows converting deposit ↔ rent |
| deposit_m, rent_m | float | deposit and monthly rent as asked |
| published_date | date | Divar's publish date |
| first_seen, last_seen | date | first and last crawl snapshot that listed the ad |
| days_listed | int | `last_seen` − `published_date` |
| group | str | hood, build year and floor joined by `/`, for grouped CV |
| equiv_m | float | full-deposit equivalent of the flat |
| ppm | float | `equiv_m / area_m2` |
| log_ppm | float | target, `ln(ppm)` |

## Collection and cleaning

Each weekly crawl reads the newest 150 rent ads in every neighborhood. An ad seen in several snapshots appears once,
as its latest version. Within each zone, the cleaning drops ads with an area outside the zone's size range, ads missing
the area, build year or price, and ads for shared housing, offices, furnished or short-term lets, or a single room. It
also drops ads whose price converts to zero, repeats of another ad's area, build year, floor, deposit and rent (the
newest copy stays), and ads more than 3 MADs from the zone's median log price per m².

| Zone | Name | Size m² | Ads | Rows | Dropped |
|---|---|---|---|---|---|
| west | غرب تهران | 40 to 110 | 19,813 | 15,511 | out_of_size 865, missing 24, not_whole_flat 1,503, zero 7, dup 1,611, outlier 292 |
| east | شرق تهران | 40 to 110 | 11,590 | 9,780 | out_of_size 191, missing 17, not_whole_flat 761, zero 7, dup 765, outlier 69 |
| north | شمال تهران | 40 to 110 | 4,053 | 2,976 | out_of_size 390, missing 8, not_whole_flat 363, zero 3, dup 272, outlier 41 |
| center | مرکز تهران | 40 to 110 | 8,749 | 6,579 | out_of_size 198, missing 27, not_whole_flat 1,266, zero 3, dup 603, outlier 73 |
| south | جنوب تهران | 40 to 110 | 4,235 | 3,485 | out_of_size 90, missing 2, not_whole_flat 382, zero 8, dup 228, outlier 40 |
| karaj | کرج | 40 to 110 | 5,830 | 4,602 | out_of_size 357, missing 7, not_whole_flat 400, zero 5, dup 425, outlier 34 |

The dataset leaves out ad titles, descriptions and map pins.

## Baseline

The site's own model is an OLS with a neighborhood fixed effect. Its inputs are age (with hinges at 3, 8, 15 and 25
years), log size, parking, elevator, flights climbed without a lift, renovated, complex and days since publishing. On
the forward week ending 2026-10-13 in west Tehran, its median absolute percentage error was about 9%. The method is
described at [rent.nlogn.ir](https://rent.nlogn.ir/#method).

## License

Released under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/). Each row is the structured fields of a
public Divar ad, linked in `url`. Attribute as "Tehran rent asks, rent.nlogn.ir, from Divar listings".

	---
	license: cc-by-sa-4.0
	language:
	- fa
	- en
	pretty_name: Tehran & Karaj apartment rent asks (Divar)
	size_categories:
	- 10K<n<100K
	task_categories:
	- tabular-regression
	tags:
	- real-estate
	- rent
	- tehran
	- iran
	- tabular
	configs:
	- config_name: default
	data_files:
	- split: train
	path: data/train.parquet
	- split: test
	path: data/test.parquet
	---

	# Tehran & Karaj apartment rent asks (Divar)

	Asking rents for whole flats on yearly leases, listed on [Divar](https://divar.ir) in 493 neighborhoods of Tehran
	and Karaj. The Tehran part covers all 22 municipal districts, split into five zones. The data comes from the weekly
	crawls behind [rent.nlogn.ir](https://rent.nlogn.ir), snapshots 2026-10-09, 2026-10-13, 2026-10-15, 2026-10-19, 2026-10-20, 2026-10-22, 2026-10-27.

	These are asking prices from the ads. Signed leases usually close somewhat lower. All money is in million toman.

	## Task

	The target is `log_ppm`, the natural log of the price per m² after converting the ad to a full deposit (رهن کامل):
	`ppm = (deposit_m + rent_m / 0.03) / area_m2`. The 3% monthly rate is the median that Divar
	posters set on the convertible-rent slider. `equiv_m` is the same price for the whole flat.

	`deposit_m`, `rent_m`, `equiv_m` and `ppm` restate the target in other units, so leave them out of the inputs.

	## Splits

	The split is forward in time by publish date. `test` holds the ads published in the last 7 days,
	after 2026-10-20 up to 2026-10-27, and `train` holds everything published on or before 2026-10-20.

	\| Split \| Rows \|
	\|---\|---\|
	\| train \| 33,501 \|
	\| test \| 9,432 \|

	Agencies often repost the same flat. For cross-validation on `train`, group folds by `group` (neighborhood, build year
	and floor) so the copies stay in one fold.

	## Columns

	\| Column \| Type \| Meaning \|
	\|---\|---\|---\|
	\| token, url \| str \| Divar ad id and link \|
	\| city, zone \| str \| `tehran`/`karaj`; site zone (west, east, north, center, south, karaj) \|
	\| hood, hood_name \| str \| Divar neighborhood slug and Persian name \|
	\| hood_lat, hood_lon \| float \| Divar's public centroid of the neighborhood (not the ad's pin) \|
	\| area_m2 \| int \| floor area \|
	\| year_built \| int \| Jalali build year; Divar caps the oldest at 1370 ("1370 or before") \|
	\| age \| int \| Jalali year at `last_seen` − `year_built` \|
	\| floor, floors_total, rooms \| int \| floor (0 = ground; may be null), building floors (often null), bedrooms \|
	\| elevator, parking, storage \| bool \| Divar's structured amenities \|
	\| renovated, complex \| bool \| ad text mentions renovation / a complex (مجتمع، شهرک، برج …) \|
	\| convertible \| bool \| the poster allows converting deposit ↔ rent \|
	\| deposit_m, rent_m \| float \| deposit and monthly rent as asked \|
	\| published_date \| date \| Divar's publish date \|
	\| first_seen, last_seen \| date \| first and last crawl snapshot that listed the ad \|
	\| days_listed \| int \| `last_seen` − `published_date` \|
	\| group \| str \| hood, build year and floor joined by `/`, for grouped CV \|
	\| equiv_m \| float \| full-deposit equivalent of the flat \|
	\| ppm \| float \| `equiv_m / area_m2` \|
	\| log_ppm \| float \| target, `ln(ppm)` \|

	## Collection and cleaning

	Each weekly crawl reads the newest 150 rent ads in every neighborhood. An ad seen in several snapshots appears once,
	as its latest version. Within each zone, the cleaning drops ads with an area outside the zone's size range, ads missing
	the area, build year or price, and ads for shared housing, offices, furnished or short-term lets, or a single room. It
	also drops ads whose price converts to zero, repeats of another ad's area, build year, floor, deposit and rent (the
	newest copy stays), and ads more than 3 MADs from the zone's median log price per m².

	\| Zone \| Name \| Size m² \| Ads \| Rows \| Dropped \|
	\|---\|---\|---\|---\|---\|---\|
	\| west \| غرب تهران \| 40 to 110 \| 19,813 \| 15,511 \| out_of_size 865, missing 24, not_whole_flat 1,503, zero 7, dup 1,611, outlier 292 \|
	\| east \| شرق تهران \| 40 to 110 \| 11,590 \| 9,780 \| out_of_size 191, missing 17, not_whole_flat 761, zero 7, dup 765, outlier 69 \|
	\| north \| شمال تهران \| 40 to 110 \| 4,053 \| 2,976 \| out_of_size 390, missing 8, not_whole_flat 363, zero 3, dup 272, outlier 41 \|
	\| center \| مرکز تهران \| 40 to 110 \| 8,749 \| 6,579 \| out_of_size 198, missing 27, not_whole_flat 1,266, zero 3, dup 603, outlier 73 \|
	\| south \| جنوب تهران \| 40 to 110 \| 4,235 \| 3,485 \| out_of_size 90, missing 2, not_whole_flat 382, zero 8, dup 228, outlier 40 \|
	\| karaj \| کرج \| 40 to 110 \| 5,830 \| 4,602 \| out_of_size 357, missing 7, not_whole_flat 400, zero 5, dup 425, outlier 34 \|

	The dataset leaves out ad titles, descriptions and map pins.

	## Baseline

	The site's own model is an OLS with a neighborhood fixed effect. Its inputs are age (with hinges at 3, 8, 15 and 25
	years), log size, parking, elevator, flights climbed without a lift, renovated, complex and days since publishing. On
	the forward week ending 2026-10-13 in west Tehran, its median absolute percentage error was about 9%. The method is
	described at [rent.nlogn.ir](https://rent.nlogn.ir/#method).

	## License

	Released under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/). Each row is the structured fields of a
	public Divar ad, linked in `url`. Attribute as "Tehran rent asks, rent.nlogn.ir, from Divar listings".