In [ ]:

    

%load_ext autoreload
%autoreload 2


    

In [ ]:

    

%matplotlib inline

from fastai.imports import *
from fastai.torch_imports import *
from fastai.dataset import *
from fastai.learner import *
from fastai.structured import *
from fastai.column_data import *


    

In [ ]:

    

dep = 'SalePrice'
PATH = "data/bulldozers/"
df_raw = pd.read_feather('tmp/raw')
keep_cols = list(np.load('tmp/keep_cols.npy'))


    

In [ ]:

    

df_raw.loc[df_raw.YearMade<1950, 'YearMade'] = 1950
df_raw['age'] = df_raw.saleYear-df_raw.YearMade
df_raw = df_raw[keep_cols+['age', dep]].copy()
df_indep = df_raw.drop(dep,axis=1)

n_valid = 12000
n_trn = len(df_raw)-n_valid


    

In [ ]:

    

cat_flds = [n for n in df_indep.columns if df_raw[n].nunique()<n_trn/50]
' '.join(cat_flds)


    

In [ ]:

    

for o in ['saleElapsed', 'saleDayofyear', 'saleDay', 'age', 'YearMade']: cat_flds.remove(o)
[n for n in df_indep.drop(cat_flds,axis=1).columns if not is_numeric_dtype(df_raw[n])]


    

In [ ]:

    

for n in cat_flds: df_raw[n] = df_raw[n].astype('category').cat.as_ordered()

cont_flds = [n for n in df_indep.columns if n not in cat_flds]
' '.join(cont_flds)


    

In [ ]:

    

df_raw = df_raw[cat_flds+cont_flds+[dep]]
df, y, mapper = proc_df(df_raw, 'SalePrice', do_scale=True)

val_idx = list(range(n_trn, len(df)))


    

In [ ]:

    

md = ColumnarModelData.from_data_frame(PATH, val_idx, df, y, cat_flds=cat_flds, bs=64)


    

In [ ]:

    

df.head()


    

In [ ]:

    

def rmse(x,y): return math.sqrt(((x-y)**2).mean())


    

In [ ]:

    

emb_c = {n: len(c.cat.categories)+1 for n,c in df_raw[cat_flds].items()}
emb_c


    

In [ ]:

    

emb_szs = [(c, min(50, (c+1)//2)) for _,c in emb_c.items()]
metrics=[rmse]


    

In [ ]:

    

y_range=(0,np.max(y)*1.2)


    

In [ ]:

    

m = md.get_learner(emb_szs, len(cont_flds), 0.05, 1, [500,250], [0.5,0.05],
                   y_range=y_range, use_bn=True)


    

In [ ]:

    

m.lr_find()


    

In [ ]:

    

m.sched.plot(1300)


    

In [ ]:

    

lr=1e-3; wd=1e-7


    

In [ ]:

    

m.fit(lr, 2, wd, cycle_len=1, cycle_mult=2)


    

In [ ]:

    

m.fit(lr, 2, wd, cycle_len=2, cycle_mult=2)


    

In [ ]:

    

math.sqrt(0.0487)