0

目的是通过 gridsearchCV 优化 Keras LSTM 的超参数。第一次训练后会产生维度错误。RNN-LSTM 需要 3D 输入。gridsearchCV 是否存在维度冲突？这里的代码：

def createLSTMModel(dropout_rate=0.1, optimizer='Nadam', learning_rate=0.015, activation='relu', loss='mae', n_jobs=1):

    K.clear_session()

    model = Sequential()
    model.add(Bidirectional(LSTM(250, return_sequences=True),input_shape=(Train_Num,1)))
    model.add(Dropout(dropout_rate))
    model.add(Bidirectional(LSTM(170, return_sequences=True)))
    model.add(Dropout(dropout_rate))
    model.add(Bidirectional(LSTM(25, return_sequences=True)))
    model.add(Dropout(dropout_rate))
    #model.add(Flatten())
    model.add(Dense(1))
    model.compile(optimizer=optimizer,loss=loss)
    return model

grid_param_LSTM = {
    'batch_size': [348,256,200,128],
    'epochs': [15,30],   
    'learning_rate':[0.001,0.01,0.1,0.0001],
    'optimizer': ['Nadam', 'Adam', 'RMSProp'],
    'loss': ['logcosh', 'mae', 'mse', 'hinge','squared_hinge'],
    'activation': ['relu', 'linear','sigmoid','hard_sigmoid', 'tanh'],
    'dropout_rate':[0.1,0.2,0.4,0.6]
}

model_LSTM=KerasRegressor(build_fn=createLSTMModel)

GridLSTM = GridSearchCV(estimator=model_LSTM,
                     param_grid=grid_param_LSTM,
                     scoring={'neg_mean_squared_error','r2','explained_variance','max_error','neg_mean_absolute_error','neg_median_absolute_error'},
                     refit='neg_median_absolute_error',
                     cv=2)

X_train_R_Grid=X_train_smallLSTM.reshape(4,Train_Num,1)
y_train_R_Grid=y_train_smallLSTM.reshape(4,Train_Num,1)

GridLSTM.fit(X_train_R_Grid, y_train_R_Grid)

产生以下错误：

Epoch 1/15
1/1 [==============================] - 17s 17s/step - loss: 1.4543
Epoch 2/15
1/1 [==============================] - 15s 15s/step - loss: 1.4568
Epoch 3/15
1/1 [==============================] - 13s 13s/step - loss: 1.3417
Epoch 4/15
1/1 [==============================] - 13s 13s/step - loss: 1.2683
Epoch 5/15
1/1 [==============================] - 14s 14s/step - loss: 1.2689
Epoch 6/15
1/1 [==============================] - 13s 13s/step - loss: 1.2692
Epoch 7/15
1/1 [==============================] - 12s 12s/step - loss: 1.2667
Epoch 8/15
1/1 [==============================] - 12s 12s/step - loss: 1.2663
Epoch 9/15
1/1 [==============================] - 12s 12s/step - loss: 1.2685
Epoch 10/15
1/1 [==============================] - 12s 12s/step - loss: 1.2660
Epoch 11/15
1/1 [==============================] - 12s 12s/step - loss: 1.2628
Epoch 12/15
1/1 [==============================] - 14s 14s/step - loss: 1.2666
Epoch 13/15
1/1 [==============================] - 13s 13s/step - loss: 1.2662
Epoch 14/15
1/1 [==============================] - 13s 13s/step - loss: 1.2656
Epoch 15/15
1/1 [==============================] - 13s 13s/step - loss: 1.2644

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-251-ec253ff8a2db> in <module>
----> 1 GridLSTM.fit(X_train_R_Grid, y_train_R_Grid)

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\model_selection\_search.py in fit(self, X, y, groups, **fit_params)
    685                 return results
    686 
--> 687             self._run_search(evaluate_candidates)
    688 
    689         # For multi-metric evaluation, store the best_index_, best_params_ and

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\model_selection\_search.py in _run_search(self, evaluate_candidates)
   1146     def _run_search(self, evaluate_candidates):
   1147         """Search all candidates in param_grid"""
-> 1148         evaluate_candidates(ParameterGrid(self.param_grid))
   1149 
   1150 

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\model_selection\_search.py in evaluate_candidates(candidate_params)
    664                                for parameters, (train, test)
    665                                in product(candidate_params,
--> 666                                           cv.split(X, y, groups)))
    667 
    668                 if len(out) < 1:

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\parallel.py in __call__(self, iterable)
    919             # remaining jobs.
    920             self._iterating = False
--> 921             if self.dispatch_one_batch(iterator):
    922                 self._iterating = self._original_iterator is not None
    923 

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\parallel.py in dispatch_one_batch(self, iterator)
    757                 return False
    758             else:
--> 759                 self._dispatch(tasks)
    760                 return True
    761 

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\parallel.py in _dispatch(self, batch)
    714         with self._lock:
    715             job_idx = len(self._jobs)
--> 716             job = self._backend.apply_async(batch, callback=cb)
    717             # A job can complete so quickly than its callback is
    718             # called before we get here, causing self._jobs to

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\_parallel_backends.py in apply_async(self, func, callback)
    180     def apply_async(self, func, callback=None):
    181         """Schedule a func to be run"""
--> 182         result = ImmediateResult(func)
    183         if callback:
    184             callback(result)

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\_parallel_backends.py in __init__(self, batch)
    547         # Don't delay the application, to avoid keeping the input
    548         # arguments in memory
--> 549         self.results = batch()
    550 
    551     def get(self):

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\parallel.py in __call__(self)
    223         with parallel_backend(self._backend, n_jobs=self._n_jobs):
    224             return [func(*args, **kwargs)
--> 225                     for func, args, kwargs in self.items]
    226 
    227     def __len__(self):

~\Anaconda3\envs\Tensorflow\lib\site-packages\joblib\parallel.py in <listcomp>(.0)
    223         with parallel_backend(self._backend, n_jobs=self._n_jobs):
    224             return [func(*args, **kwargs)
--> 225                     for func, args, kwargs in self.items]
    226 
    227     def __len__(self):

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\model_selection\_validation.py in _fit_and_score(estimator, X, y, scorer, train, test, verbose, parameters, fit_params, return_train_score, return_parameters, return_n_test_samples, return_times, return_estimator, error_score)
    552         fit_time = time.time() - start_time
    553         # _score will return dict if is_multimetric is True
--> 554         test_scores = _score(estimator, X_test, y_test, scorer, is_multimetric)
    555         score_time = time.time() - start_time - fit_time
    556         if return_train_score:

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\model_selection\_validation.py in _score(estimator, X_test, y_test, scorer, is_multimetric)
    595     """
    596     if is_multimetric:
--> 597         return _multimetric_score(estimator, X_test, y_test, scorer)
    598     else:
    599         if y_test is None:

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\model_selection\_validation.py in _multimetric_score(estimator, X_test, y_test, scorers)
    625             score = scorer(estimator, X_test)
    626         else:
--> 627             score = scorer(estimator, X_test, y_test)
    628 
    629         if hasattr(score, 'item'):

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\metrics\scorer.py in __call__(self, estimator, X, y_true, sample_weight)
     95         else:
     96             return self._sign * self._score_func(y_true, y_pred,
---> 97                                                  **self._kwargs)
     98 
     99 

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\metrics\regression.py in explained_variance_score(y_true, y_pred, sample_weight, multioutput)
    413     """
    414     y_type, y_true, y_pred, multioutput = _check_reg_targets(
--> 415         y_true, y_pred, multioutput)
    416     check_consistent_length(y_true, y_pred, sample_weight)
    417 

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\metrics\regression.py in _check_reg_targets(y_true, y_pred, multioutput)
     75 
     76     """
---> 77     check_consistent_length(y_true, y_pred)
     78     y_true = check_array(y_true, ensure_2d=False)
     79     y_pred = check_array(y_pred, ensure_2d=False)

~\Anaconda3\envs\Tensorflow\lib\site-packages\sklearn\utils\validation.py in check_consistent_length(*arrays)
    203     if len(uniques) > 1:
    204         raise ValueError("Found input variables with inconsistent numbers of"
--> 205                          " samples: %r" % [int(l) for l in lengths])
    206 
    207 

ValueError: Found input variables with inconsistent numbers of samples: [1, 1396]

软件包版本

System:
    python: 3.6.9 |Anaconda, Inc.| (default, Jul 30 2019, 14:00:49) [MSC v.1915 64 bit (AMD64)]
executable: C:\Users\alias\Anaconda3\envs\Tensorflow\python.exe
   machine: Windows-10-10.0.18362-SP0

BLAS:
    macros: 
  lib_dirs: 
cblas_libs: cblas

Python deps:
       pip: 19.1.1
setuptools: 39.1.0
   sklearn: 0.21.2
     numpy: 1.15.0
     scipy: 1.2.1
    Cython: 0.28.4
    pandas: 0.24.2
keras: 2.1.6
tensorflow:1.9.0

keras、numpy 和 tensorflow 版本是特意选择来实现K.clear_session()的，它结合了gridsearchCV最新版本的 OOM 问题。

gridsearchCV 不适合 RNN 使用还是有解决方法？

提前非常感谢。

最好的问候， JDS

4

2 回答 2

0

问题是得分。带有维度转换的自定义记分器提供了解决方案。就这么简单：

def CustomMAD(y_true, y_pred):
    y_true=y_true.reshape(Train_Num,1)

    if np.isnan(y_pred).any():
        result=-1000000
    else:
        y_pred=y_pred.reshape(Train_Num,1)
        result=median_absolute_error(y_true, y_pred)

    return result

scorer = make_scorer(CustomMAD, greater_is_better=False)

GridLSTM = GridSearchCV(model_LSTM,
                     param_grid=grid_param_LSTM,
                     scoring=scorer,
                     cv=3)
于 2019-08-08T14:06:38.413 回答
0

是的。GridSearch需要二维数组作为输入，因此在您的情况下将失败。

另请参阅：带有 LSTM 的 GridSearchCV/RandomizedSearchCV

作为一个工作示例，请参阅以下内容。使用 2D 输入的 keras 和 GridSearch 的组合。

grid.fit(X, Y)正在使用X它需要是一个二维数组。

数据集：https ://raw.githubusercontent.com/jbrownlee/Datasets/master/pima-indians-diabetes.data.csv

# Use scikit-learn to grid search the batch size and epochs
import numpy
from sklearn.model_selection import GridSearchCV
from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasClassifier
# Function to create model, required for KerasClassifier
def create_model():
    # create model
    model = Sequential()
    model.add(Dense(12, input_dim=8, activation='relu'))
    model.add(Dense(1, activation='sigmoid'))
    # Compile model
    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
    return model
# fix random seed for reproducibility
seed = 7
numpy.random.seed(seed)
# load dataset
dataset = numpy.loadtxt("pima-indians-diabetes.csv", delimiter=",")
# split into input (X) and output (Y) variables
X = dataset[:,0:8]
Y = dataset[:,8]
# create model
model = KerasClassifier(build_fn=create_model, verbose=0)
# define the grid search parameters
batch_size = [10, 20, 40, 60, 80, 100]
epochs = [10, 50, 100]
param_grid = dict(batch_size=batch_size, epochs=epochs)
grid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=-1)
grid_result = grid.fit(X, Y)
# summarize results
print("Best: %f using %s" % (grid_result.best_score_, grid_result.best_params_))
means = grid_result.cv_results_['mean_test_score']
stds = grid_result.cv_results_['std_test_score']
params = grid_result.cv_results_['params']
for mean, stdev, param in zip(means, stds, params):
    print("%f (%f) with: %r" % (mean, stdev, param))
于 2019-08-05T20:06:24.063 回答