In [1]:

    

%pylab inline
import pandas as pd


    

    




Populating the interactive namespace from numpy and matplotlib

In [26]:

    

def rmsle(actual, predicted):
    error = np.log1p(predicted) - np.log1p(actual)
    return np.sqrt(np.mean(np.square(error)))


    

In [9]:

    

train_set = pd.read_csv('../train_set_ym.csv')


    

In [10]:

    

train_set.dtypes


    

    
Out[10]:





tube_assembly_id       object
supplier               object
annual_usage            int64
min_order_quantity      int64
bracket_pricing        object
quantity                int64
cost                  float64
year                    int64
month                   int64
dtype: object

In [305]:

    

model_data = train_set[train_set.bracket_pricing == 'No']


    

In [306]:

    

model_data = model_data.drop(['bracket_pricing'], axis=1)


    

In [289]:

    

model_data = pd.get_dummies(model_data, columns=['supplier'])


    

In [307]:

    

tube = pd.read_csv('../tube_material_id_imputed_dummies_drop_ns.csv')


    

In [308]:

    

comp_weight = pd.read_csv('../comp_weight.csv')


    

In [309]:

    

# model_data = pd.merge(model_data, tube, how='left', on='tube_assembly_id')
model_data = pd.merge(model_data, comp_weight, how='left', on='tube_assembly_id')


    

In [310]:

    

print model_data.shape
model_data.head()


    

    




(3930, 9)






    
Out[310]:






  

    

      

      
tube_assembly_id
      
supplier
      
annual_usage
      
min_order_quantity
      
quantity
      
cost
      
year
      
month
      
weight
    
  
  

    

      
0
      
TA-00013
      
S-0026
      
50
      
1
      
1
      
10.004284
      
1992
      
7
      
0.215
    
    

      
1
      
TA-00021
      
S-0030
      
1
      
1
      
1
      
3.433061
      
2014
      
2
      
0.052
    
    

      
2
      
TA-00022
      
S-0013
      
1
      
1
      
1
      
8.563178
      
2011
      
7
      
0.018
    
    

      
3
      
TA-00027
      
S-0062
      
30
      
1
      
1
      
2.737546
      
1993
      
9
      
0.028
    
    

      
4
      
TA-00037
      
S-0062
      
600
      
1
      
1
      
1.563517
      
1994
      
9
      
0.024
    
  

In [224]:

    

import sklearn.cross_validation as skcv
import sklearn.ensemble as sken
import sklearn.grid_search as skgs


    

In [311]:

    

model_train, model_test = skcv.train_test_split(model_data, random_state=5, train_size=0.7)


    

In [283]:

    

# skcv.KFold()


    

In [312]:

    

print model_train.shape
print model_test.shape


    

    




(2751, 9)
(1179, 9)

In [295]:

    

rf = sken.RandomForestRegressor(n_estimators=100, max_features=5)
# rf = sken.ExtraTreesRegressor(n_estimators=50)
# rf = sken.GradientBoostingRegressor(learning_rate=0.25, n_estimators=200)


    

In [313]:

    

model_train_X = model_train.drop(['tube_assembly_id', 'cost'], axis=1)
rf = rf.fit(model_train_X, np.log1p(model_train.cost))


    

    




---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-313-11f07db00579> in <module>()
      1 model_train_X = model_train.drop(['tube_assembly_id', 'cost'], axis=1)
----> 2 rf = rf.fit(model_train_X, np.log1p(model_train.cost))

C:\Users\shenbe1\AppData\Local\Continuum\Anaconda\lib\site-packages\sklearn\ensemble\forest.pyc in fit(self, X, y, sample_weight)
    193         """
    194         # Validate or convert input data
--> 195         X = check_array(X, dtype=DTYPE, accept_sparse="csc")
    196         if issparse(X):
    197             # Pre-sort indices to avoid that each individual tree of the

C:\Users\shenbe1\AppData\Local\Continuum\Anaconda\lib\site-packages\sklearn\utils\validation.pyc in check_array(array, accept_sparse, dtype, order, copy, force_all_finite, ensure_2d, allow_nd, ensure_min_samples, ensure_min_features)
    342             else:
    343                 dtype = None
--> 344         array = np.array(array, dtype=dtype, order=order, copy=copy)
    345         # make sure we actually converted to numeric:
    346         if dtype_numeric and array.dtype.kind == "O":

ValueError: could not convert string to float: S-0013

In [297]:

    

model_train_pred = np.expm1(rf.predict(model_train_X))
print rmsle(model_train.cost, model_train_pred)


    

    




0.215507346458

In [298]:

    

model_train_actual_fit = pd.DataFrame({'actual' : log1p(model_train.cost), 'fit' : log1p(model_train_pred)})


    

In [299]:

    

model_feature_importance = pd.DataFrame({'feature' : model_train_X.columns, 'score' : rf.feature_importances_})


    

In [304]:

    

model_feature_importance


    

    
Out[304]:






  

    

      

      
feature
      
score
    
  
  

    

      
0
      
annual_usage
      
0.217973
    
    

      
1
      
min_order_quantity
      
0.066302
    
    

      
2
      
quantity
      
0.051963
    
    

      
3
      
year
      
0.077305
    
    

      
4
      
month
      
0.068600
    
    

      
5
      
supplier_S-0003
      
0.001244
    
    

      
6
      
supplier_S-0004
      
0.000085
    
    

      
7
      
supplier_S-0005
      
0.001694
    
    

      
8
      
supplier_S-0006
      
0.000186
    
    

      
9
      
supplier_S-0007
      
0.001097
    
    

      
10
      
supplier_S-0009
      
0.001860
    
    

      
11
      
supplier_S-0011
      
0.000229
    
    

      
12
      
supplier_S-0012
      
0.000360
    
    

      
13
      
supplier_S-0013
      
0.015897
    
    

      
14
      
supplier_S-0014
      
0.001260
    
    

      
15
      
supplier_S-0015
      
0.000396
    
    

      
16
      
supplier_S-0022
      
0.000000
    
    

      
17
      
supplier_S-0023
      
0.000432
    
    

      
18
      
supplier_S-0024
      
0.000226
    
    

      
19
      
supplier_S-0025
      
0.000082
    
    

      
20
      
supplier_S-0026
      
0.043541
    
    

      
21
      
supplier_S-0027
      
0.001861
    
    

      
22
      
supplier_S-0029
      
0.000090
    
    

      
23
      
supplier_S-0030
      
0.007378
    
    

      
24
      
supplier_S-0031
      
0.000671
    
    

      
25
      
supplier_S-0041
      
0.046802
    
    

      
26
      
supplier_S-0042
      
0.000343
    
    

      
27
      
supplier_S-0043
      
0.000535
    
    

      
28
      
supplier_S-0046
      
0.000166
    
    

      
29
      
supplier_S-0051
      
0.000298
    
    

      
30
      
supplier_S-0054
      
0.000302
    
    

      
31
      
supplier_S-0058
      
0.004721
    
    

      
32
      
supplier_S-0059
      
0.000407
    
    

      
33
      
supplier_S-0060
      
0.000783
    
    

      
34
      
supplier_S-0061
      
0.000000
    
    

      
35
      
supplier_S-0062
      
0.006985
    
    

      
36
      
supplier_S-0064
      
0.006395
    
    

      
37
      
supplier_S-0066
      
0.078132
    
    

      
38
      
supplier_S-0068
      
0.000000
    
    

      
39
      
supplier_S-0070
      
0.000621
    
    

      
40
      
supplier_S-0072
      
0.021681
    
    

      
41
      
supplier_S-0074
      
0.000000
    
    

      
42
      
supplier_S-0078
      
0.000772
    
    

      
43
      
supplier_S-0081
      
0.002939
    
    

      
44
      
supplier_S-0087
      
0.000000
    
    

      
45
      
supplier_S-0092
      
0.000790
    
    

      
46
      
supplier_S-0095
      
0.000146
    
    

      
47
      
supplier_S-0096
      
0.000121
    
    

      
48
      
supplier_S-0104
      
0.006360
    
    

      
49
      
supplier_S-0105
      
0.003140
    
    

      
50
      
supplier_S-0106
      
0.002206
    
    

      
51
      
supplier_S-0107
      
0.000312
    
    

      
52
      
supplier_S-0108
      
0.000045
    
    

      
53
      
supplier_S-0109
      
0.000143
    
    

      
54
      
supplier_S-0111
      
0.000031
    
    

      
55
      
weight
      
0.254089
    
  

In [300]:

    

important_feats = model_feature_importance.sort(columns=['score'], ascending=False).head(12).feature.values


    

In [303]:

    

important_feats


    

    
Out[303]:





array(['weight', 'annual_usage', 'supplier_S-0066', 'year', 'month',
       'min_order_quantity', 'quantity', 'supplier_S-0041',
       'supplier_S-0026', 'supplier_S-0072', 'supplier_S-0013',
       'supplier_S-0030'], dtype=object)

In [301]:

    

model_train_actual_fit.plot(x='actual', y='fit', kind='scatter')
plt.show()


    

In [302]:

    

model_test_X = model_test.drop(['tube_assembly_id', 'cost'], axis=1)
model_test_pred = np.expm1(rf.predict(model_test_X))
print rmsle(model_test.cost, model_test_pred)


    

    




0.533998807167

In [268]:

    

model_test_actual_fit = pd.DataFrame({'actual' : log1p(model_test.cost), 'fit' : log1p(model_test_pred)})


    

In [269]:

    

model_test_actual_fit.plot(x='actual', y='fit', kind='scatter')
plt.show()


    

In [270]:

    

model_train_X_imps = model_train[important_feats]
rf = rf.fit(model_train_X_imps, np.log1p(model_train.cost))


    

In [271]:

    

model_train_imps_pred = np.expm1(rf.predict(model_train_X_imps))
print rmsle(model_train.cost, model_train_imps_pred)


    

    




0.202504420185

In [272]:

    

model_test_X = model_test.drop(['tube_assembly_id', 'cost'], axis=1)
model_test_X_imps = model_test_X[important_feats]
model_test_pred = np.expm1(rf.predict(model_test_X_imps))
print rmsle(model_test.cost, model_test_pred)


    

    




0.53914474706

In [273]:

    

rf = sken.RandomForestRegressor(n_estimators=100, max_features=5, random_state=0)
gridcv = skgs.GridSearchCV(rf, param_grid={'n_estimators' : [50, 100, 200, 500]})


    

In [274]:

    

gridcv.fit(model_train_X, log1p(model_train.cost))


    

    
Out[274]:





GridSearchCV(cv=None, error_score='raise',
       estimator=RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=None,
           max_features=5, max_leaf_nodes=None, min_samples_leaf=1,
           min_samples_split=2, min_weight_fraction_leaf=0.0,
           n_estimators=100, n_jobs=1, oob_score=False, random_state=0,
           verbose=0, warm_start=False),
       fit_params={}, iid=True, loss_func=None, n_jobs=1,
       param_grid={'n_estimators': [50, 100, 200, 500]},
       pre_dispatch='2*n_jobs', refit=True, score_func=None, scoring=None,
       verbose=0)

In [275]:

    

gridcv.best_estimator_


    

    
Out[275]:





RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=None,
           max_features=5, max_leaf_nodes=None, min_samples_leaf=1,
           min_samples_split=2, min_weight_fraction_leaf=0.0,
           n_estimators=500, n_jobs=1, oob_score=False, random_state=0,
           verbose=0, warm_start=False)

In [276]:

    

model_train_pred = expm1(gridcv.predict(model_train_X))
print rmsle(model_train.cost, model_train_pred)


    

    




0.196731086559

In [277]:

    

model_test_X = model_test.drop(['tube_assembly_id', 'cost'], axis=1)
model_test_pred = expm1(gridcv.predict(model_test_X))
print rmsle(model_test.cost, model_test_pred)


    

    




0.532419121412