'rsh_estimate' returns '0's for coefficient values

UDST / sanfran_urbansim

An UrbanSim for San Francisco: an example implementation of the new framework

39 stars 27 forks source link

'rsh_estimate' returns '0's for coefficient values #12

Closed lisalan520 closed 9 years ago

lisalan520 commented 9 years ago

Hi,

I got '0's for all coefficients when I ran sim.run(["rsh_estimate"]) in Estimation.ipynb. And it re-wrote my rsh.yaml file. The Hedonic Example notebook returned 'correct'-looking values with directly running the fit function. Below are two captures. Have you got any similar issues? Thanks!

Estimation.ipynb capture

Hedonic Example.ipynb capture

fscottfoti commented 9 years ago

Not entirely sure - there might be a filter that is removing most/all of the records?

lisalan520 commented 9 years ago

It is a little odd.. as I didn't change any thing. The fit filters in the rsh.yaml are:

fit_filters:

unit_lot_size > 0
year_built > 1000
year_built < 2020
unit_sqft > 100
unit_sqft < 20000

which is the same as is in Hedonic Example...

lisalan520 commented 9 years ago

Hi @fscottfoti, Just solved the problem by removing line 20-23 in dataset.py. They are the 'filters' you mentioned. Thanks!

fscottfoti commented 9 years ago

Great! I suspect there was no variability in one or more columns when the filters removed enough data. We should probably be detecting that error directly.