In [19]:

    

###PATH = "C:/Users/magaxels/AutoML"
PATH = "/Users/magnusax/AutoML/"
import sys; sys.path.insert(0, PATH)
from gazer import GazerMetaLearner
from gazer.ensembler import GazerMetaEnsembler
import numpy as np
import pandas as pd
import os
import time
from gazer.metrics import get_scorer
from sklearn.externals import joblib
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_digits
import random


    

In [20]:

    

X, y = load_digits(return_X_y=True)


    

In [21]:

    

X.shape, y.shape


    

    
Out[21]:





((1797, 64), (1797,))

In [22]:

    

X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.5, random_state=0)
print(X_train.shape, y_train.shape, X_val.shape, y_val.shape)


    

    




(898, 64) (898,) (899, 64) (899,)

In [23]:

    

from sklearn.preprocessing import StandardScaler, MaxAbsScaler
##scaler = StandardScaler()
#scaler = StandardScaler()
#scaler.fit(X_train)
#X_train = scaler.transform(X_train)
#X_val = scaler.transform(X_val)


    

In [24]:

    

learner = GazerMetaLearner(method='all', 
                           estimators=['adaboost', 'multinomial_nb', 'logreg', 'knn', 'neuralnet'], 
                           verbose=1)
learner.update('neuralnet', {'n_hidden': 3, 'epochs': 50, 'input_units': 1000, 'p': 0.1, 'batch_norm': True})


    

    




Available algorithms (use '.clf' attribute for access):
sgd_hinge, bernoulli_nb, random_forest, knn, adaboost, svm, neuralnet, multinomial_nb, tree, logreg, gaussian_nb

In [25]:

    

ensembler = GazerMetaEnsembler(learner, X_train.shape)


    

In [26]:

    

ensembler.fit(X_train, y_train, os.path.join(os.getcwd(),"tmp7"))


    

    




Training neural net..






    




neuralnet:   1%|▋                                                                       | 1/100 [00:00<00:14,  7.04it/s]





    




Train time: 0.17 min






    




neuralnet: 100%|██████████████████████████████████████████████████████████████████████| 100/100 [00:04<00:00, 22.44it/s]
sgd_hinge: 100%|██████████████████████████████████████████████████████████████████████| 200/200 [00:03<00:00, 57.37it/s]
bernoulli_nb: 100%|████████████████████████████████████████████████████████████████████| 40/40 [00:00<00:00, 207.48it/s]
multinomial_nb: 100%|██████████████████████████████████████████████████████████████████| 40/40 [00:00<00:00, 306.29it/s]
knn: 100%|██████████████████████████████████████████████████████████████████████████████| 60/60 [00:04<00:00, 12.52it/s]
gaussian_nb: 100%|████████████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 99.00it/s]
svm: 100%|████████████████████████████████████████████████████████████████████████████| 200/200 [00:08<00:00, 24.51it/s]
logreg: 100%|█████████████████████████████████████████████████████████████████████████| 200/200 [00:39<00:00,  5.13it/s]
tree: 100%|██████████████████████████████████████████████████████████████████████████| 120/120 [00:01<00:00, 111.72it/s]
adaboost: 100%|███████████████████████████████████████████████████████████████████████| 100/100 [00:01<00:00, 96.66it/s]
random_forest: 100%|██████████████████████████████████████████████████████████████████| 100/100 [09:29<00:00,  5.70s/it]

In [28]:

    

scores, ensemble = ensembler.hillclimb(X_val, y_val, 
                                       n_best=5, 
                                       p=0.5, 
                                       iterations=20, 
                                       scoring='accuracy')


    

    




ERROR:root:An unexpected error occurred while tokenizing input
The following traceback may be corrupted or invalid
The error message is: ('EOF in multi-line string', (140, 67))







    




---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-28-8e003bb7212a> in <module>()
      3                                        p=0.5,
      4                                        iterations=20,
----> 5                                        scoring='accuracy')

/Users/magnusax/AutoML/gazer/ensembler.py in hillclimb(self, X_val, y_val, n_best, p, iterations, scoring)
    356             scoring : str, default: accuracy
    357                 The metric to use when hillclimbing
--> 358 
    359         """
    360         mod_files = self._prep_output()

/Users/magnusax/AutoML/gazer/ensembler.py in _prep_output(self)
    317         and scores and return a sorted list of classifiers.
    318         """
--> 319         # External package
    320         if 'neuralnet' in self.learner.names:
    321             model, nn_fs = self.orchestrator.pop('neuralnet')

KeyError: 'neuralnet'

In [ ]:

    




    

In [ ]:

    

model, files = ensembler.orchestrator['neuralnet']


    

In [ ]:

    

files


    

In [ ]:

    

import keras
for wts, score in files:
    model.load_weights(wts)
    print(model.evaluate(X_val, keras.utils.to_categorical(y_val), verbose=0))


    

In [ ]:

    

pr = np.array([list(l).index(max(list(l))) 
               for l in model.predict(X_val)])
pr.shape


    

In [ ]:

    

pr


    

In [ ]:

    

valscores = {}; ensemble = {}
for n_best in range(1, 11):
    vs, ens = ensembler.hillclimb(X_val, y_val, n_best=n_best, p=0.8, iterations=7, scoring='accuracy')
    valscores[str(n_best)] = vs
    ensemble[str(n_best)] = ens


    

In [ ]:

    

valsscores = valscores['2']
x = [x for x,_ in valsscores]
y = [y for _,y in valsscores]
import matplotlib.pyplot as plt
% matplotlib inline
plt.figure(figsize=(12,6))
plt.plot(x,y, 'x--')
plt.show()


    

In [ ]:

    

len(ensemble['10'])


    

In [ ]: