tensorflow 연습
from sklearn.datasets import load_iris
data = load_iris()
from sklearn.model_selection import learning_curve
from sklearn.neighbors import KNeighborsClassifier
knn = KNeighborsClassifier()
from sklearn.linear_model import LogisticRegression
import numpy as np
np.linspace(0.1,1.1,10, endpoint=False)
array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1. ])
train_size, train_score, test_score = learning_curve(LogisticRegression(), data.data, data.target,
train_sizes=np.linspace(0.1,1.1,10, endpoint=False))
C:\Users\sunde\AppData\Roaming\Python\Python39\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.
Increase the number of iterations (max_iter) or scale the data as shown in:
https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
n_iter_i = _check_optimize_result(
C:\Users\sunde\AppData\Roaming\Python\Python39\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.
Increase the number of iterations (max_iter) or scale the data as shown in:
https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
n_iter_i = _check_optimize_result(
C:\Users\sunde\AppData\Roaming\Python\Python39\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.
Increase the number of iterations (max_iter) or scale the data as shown in:
https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
n_iter_i = _check_optimize_result(
C:\Users\sunde\AppData\Roaming\Python\Python39\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.
Increase the number of iterations (max_iter) or scale the data as shown in:
https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
n_iter_i = _check_optimize_result(
C:\Users\sunde\AppData\Roaming\Python\Python39\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.
Increase the number of iterations (max_iter) or scale the data as shown in:
https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
n_iter_i = _check_optimize_result(
C:\Users\sunde\AppData\Roaming\Python\Python39\site-packages\sklearn\linear_model\_logistic.py:814: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.
Increase the number of iterations (max_iter) or scale the data as shown in:
https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
n_iter_i = _check_optimize_result(
# !pip install -U sklearn-evaluation
import sklearn_evaluation
sklearn_evaluation.plot.learning_curve(train_score, test_score, train_size)
<AxesSubplot:title={'center':'Learning Curve'}, xlabel='Training examples', ylabel='Score mean'>
특징 : 전통적인 ML 기법은 타 프레임워크와 연동하게 !!!
import tensorflow as tf
(X_train,y_train),(X_test,y_test) = tf.keras.datasets.mnist.load_data()
X_train = X_train[...,tf.newaxis] / 255
X_test = X_test[...,tf.newaxis] / 255 # 안해서
y_train = tf.keras.utils.to_categorical(y_train)
import tensorflow_addons as tfa
tfa.metrics.F1Score
from sklearn.metrics import recall_score
input_ = tf.keras.Input((28,28,1))
x = tf.keras.layers.Conv2D(32,3)(input_)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.MaxPool2D()(x)
x = tf.keras.layers.Conv2D(32,3)(x)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(128)(x)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.Dense(10, activation='softmax')(x)
model = tf.keras.Model(input_, x)
model.compile(loss=tf.keras.losses.CategoricalCrossentropy(from_logits=False),
metrics=[tfa.metrics.F1Score(10)]) # F1Score 바로 구할 수 있음
# model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True), metrics=[tf.keras.metrics.Precision()])
'''
model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False), # F1Score 쓰려면 음수값 나오면 안되서 False
metrics=[tf.keras.metrics.Precision(), tf.keras.metrics.Recall()]) # Precision, Recall 구하면 F1Score 구할 수 있음
'''
model.summary()
Model: "model_52"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_52 (InputLayer) [(None, 28, 28, 1)] 0
conv2d_102 (Conv2D) (None, 26, 26, 32) 320
re_lu_153 (ReLU) (None, 26, 26, 32) 0
max_pooling2d_50 (MaxPoolin (None, 13, 13, 32) 0
g2D)
conv2d_103 (Conv2D) (None, 11, 11, 32) 9248
re_lu_154 (ReLU) (None, 11, 11, 32) 0
flatten_51 (Flatten) (None, 3872) 0
dense_102 (Dense) (None, 128) 495744
re_lu_155 (ReLU) (None, 128) 0
dense_103 (Dense) (None, 10) 1290
=================================================================
Total params: 506,602
Trainable params: 506,602
Non-trainable params: 0
_________________________________________________________________
history = model.fit(X_train,y_train,epochs=10)
Epoch 1/10
1875/1875 [==============================] - 7s 4ms/step - loss: 0.3393 - f1_score: 0.9537
Epoch 2/10
1875/1875 [==============================] - 7s 3ms/step - loss: 0.0788 - f1_score: 0.9804
Epoch 3/10
1875/1875 [==============================] - 6s 3ms/step - loss: 0.0681 - f1_score: 0.9846
Epoch 4/10
1875/1875 [==============================] - 7s 3ms/step - loss: 0.0592 - f1_score: 0.9862
Epoch 5/10
1875/1875 [==============================] - 6s 3ms/step - loss: 0.0552 - f1_score: 0.9877
Epoch 6/10
1875/1875 [==============================] - 6s 3ms/step - loss: 0.0514 - f1_score: 0.9892
Epoch 7/10
1875/1875 [==============================] - 6s 3ms/step - loss: 0.0501 - f1_score: 0.9901
Epoch 8/10
1875/1875 [==============================] - 6s 3ms/step - loss: 0.0493 - f1_score: 0.9904
Epoch 9/10
548/1875 [=======>......................] - ETA: 4s - loss: 0.0424 - f1_score: 0.9918
---------------------------------------------------------------------------
KeyboardInterrupt Traceback (most recent call last)
Input In [122], in <cell line: 1>()
----> 1 history = model.fit(X_train,y_train,epochs=10)
File C:\ProgramData\Anaconda3\lib\site-packages\keras\utils\traceback_utils.py:64, in filter_traceback.<locals>.error_handler(*args, **kwargs)
62 filtered_tb = None
63 try:
---> 64 return fn(*args, **kwargs)
65 except Exception as e: # pylint: disable=broad-except
66 filtered_tb = _process_traceback_frames(e.__traceback__)
File C:\ProgramData\Anaconda3\lib\site-packages\keras\engine\training.py:1384, in Model.fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)
1377 with tf.profiler.experimental.Trace(
1378 'train',
1379 epoch_num=epoch,
1380 step_num=step,
1381 batch_size=batch_size,
1382 _r=1):
1383 callbacks.on_train_batch_begin(step)
-> 1384 tmp_logs = self.train_function(iterator)
1385 if data_handler.should_sync:
1386 context.async_wait()
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\util\traceback_utils.py:150, in filter_traceback.<locals>.error_handler(*args, **kwargs)
148 filtered_tb = None
149 try:
--> 150 return fn(*args, **kwargs)
151 except Exception as e:
152 filtered_tb = _process_traceback_frames(e.__traceback__)
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\eager\def_function.py:915, in Function.__call__(self, *args, **kwds)
912 compiler = "xla" if self._jit_compile else "nonXla"
914 with OptionalXlaContext(self._jit_compile):
--> 915 result = self._call(*args, **kwds)
917 new_tracing_count = self.experimental_get_tracing_count()
918 without_tracing = (tracing_count == new_tracing_count)
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\eager\def_function.py:947, in Function._call(self, *args, **kwds)
944 self._lock.release()
945 # In this case we have created variables on the first call, so we run the
946 # defunned version which is guaranteed to never create variables.
--> 947 return self._stateless_fn(*args, **kwds) # pylint: disable=not-callable
948 elif self._stateful_fn is not None:
949 # Release the lock early so that multiple threads can perform the call
950 # in parallel.
951 self._lock.release()
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\eager\function.py:2956, in Function.__call__(self, *args, **kwargs)
2953 with self._lock:
2954 (graph_function,
2955 filtered_flat_args) = self._maybe_define_function(args, kwargs)
-> 2956 return graph_function._call_flat(
2957 filtered_flat_args, captured_inputs=graph_function.captured_inputs)
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\eager\function.py:1853, in ConcreteFunction._call_flat(self, args, captured_inputs, cancellation_manager)
1849 possible_gradient_type = gradients_util.PossibleTapeGradientTypes(args)
1850 if (possible_gradient_type == gradients_util.POSSIBLE_GRADIENT_TYPES_NONE
1851 and executing_eagerly):
1852 # No tape is watching; skip to running the function.
-> 1853 return self._build_call_outputs(self._inference_function.call(
1854 ctx, args, cancellation_manager=cancellation_manager))
1855 forward_backward = self._select_forward_and_backward_functions(
1856 args,
1857 possible_gradient_type,
1858 executing_eagerly)
1859 forward_function, args_with_tangents = forward_backward.forward()
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\eager\function.py:499, in _EagerDefinedFunction.call(self, ctx, args, cancellation_manager)
497 with _InterpolateFunctionError(self):
498 if cancellation_manager is None:
--> 499 outputs = execute.execute(
500 str(self.signature.name),
501 num_outputs=self._num_outputs,
502 inputs=args,
503 attrs=attrs,
504 ctx=ctx)
505 else:
506 outputs = execute.execute_with_cancellation(
507 str(self.signature.name),
508 num_outputs=self._num_outputs,
(...)
511 ctx=ctx,
512 cancellation_manager=cancellation_manager)
File C:\ProgramData\Anaconda3\lib\site-packages\tensorflow\python\eager\execute.py:54, in quick_execute(op_name, num_outputs, inputs, attrs, ctx, name)
52 try:
53 ctx.ensure_initialized()
---> 54 tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
55 inputs, attrs, num_outputs)
56 except core._NotOkStatusException as e:
57 if name is not None:
KeyboardInterrupt:
import pandas as pd
pd.DataFrame(history.history).plot.line(figsize=(10,5))
<AxesSubplot:>
pd.DataFrame(history.history).plot.line(figsize=(10,10)) # pooling (x)
<AxesSubplot:>
함수형 패러다임 >
- 반복 실험
- 함수형 연동하기 위해서
def build_fn():
input_ = tf.keras.Input((28,28,1))
x = tf.keras.layers.Conv2D(32,3)(input_)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.MaxPool2D()(x)
x = tf.keras.layers.Conv2D(32,3)(x)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(128)(x)
x = tf.keras.layers.ReLU()(x)
x = tf.keras.layers.Dense(10)(x)
model = tf.keras.Model(input_, x)
model.compile(loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
metrics=['acc'])
return model
from tensorflow.keras.wrappers.scikit_learn import KerasClassifier
model = KerasClassifier(build_fn, epochs=5) # scikit-learn 과 연동할 수 있는 텐서플로우 모델
C:\Users\sunde\AppData\Local\Temp\ipykernel_16740\964211621.py:1: DeprecationWarning: KerasClassifier is deprecated, use Sci-Keras (https://github.com/adriangb/scikeras) instead. See https://www.adriangb.com/scikeras/stable/migration.html for help migrating.
model = KerasClassifier(build_fn, epochs=5) # scikit-learn 과 연동할 수 있는 텐서플로우 모델
train_size, train_score, test_score = learning_curve(model, X_train, y_train, cv=2)
Epoch 1/5
94/94 [==============================] - 1s 5ms/step - loss: 0.6436 - acc: 0.7917
Epoch 2/5
94/94 [==============================] - 0s 3ms/step - loss: 0.1913 - acc: 0.9387
Epoch 3/5
94/94 [==============================] - 0s 3ms/step - loss: 0.1150 - acc: 0.9640
Epoch 4/5
94/94 [==============================] - 0s 3ms/step - loss: 0.0614 - acc: 0.9800
Epoch 5/5
94/94 [==============================] - 0s 3ms/step - loss: 0.0351 - acc: 0.9890
938/938 [==============================] - 2s 2ms/step - loss: 0.1415 - acc: 0.9600
94/94 [==============================] - 0s 2ms/step - loss: 0.0135 - acc: 0.9960
Epoch 1/5
305/305 [==============================] - 1s 3ms/step - loss: 0.3258 - acc: 0.8969
Epoch 2/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0934 - acc: 0.9715
Epoch 3/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0579 - acc: 0.9807
Epoch 4/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0371 - acc: 0.9881
Epoch 5/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0221 - acc: 0.9925
938/938 [==============================] - 2s 2ms/step - loss: 0.0746 - acc: 0.9792
305/305 [==============================] - 1s 2ms/step - loss: 0.0088 - acc: 0.9976
Epoch 1/5
516/516 [==============================] - 2s 3ms/step - loss: 0.2496 - acc: 0.9224
Epoch 2/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0737 - acc: 0.9767
Epoch 3/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0455 - acc: 0.9860
Epoch 4/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0305 - acc: 0.9907
Epoch 5/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0204 - acc: 0.9936
938/938 [==============================] - 2s 2ms/step - loss: 0.0622 - acc: 0.9844
516/516 [==============================] - 1s 2ms/step - loss: 0.0079 - acc: 0.9976
Epoch 1/5
727/727 [==============================] - 3s 3ms/step - loss: 0.1988 - acc: 0.9391
Epoch 2/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0660 - acc: 0.9794
Epoch 3/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0414 - acc: 0.9862
Epoch 4/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0293 - acc: 0.9919
Epoch 5/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0210 - acc: 0.9935
938/938 [==============================] - 2s 2ms/step - loss: 0.0583 - acc: 0.9844
727/727 [==============================] - 1s 2ms/step - loss: 0.0115 - acc: 0.9958
Epoch 1/5
938/938 [==============================] - 3s 3ms/step - loss: 0.1644 - acc: 0.9489
Epoch 2/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0521 - acc: 0.9832
Epoch 3/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0351 - acc: 0.9891
Epoch 4/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0255 - acc: 0.9926
Epoch 5/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0190 - acc: 0.9944
938/938 [==============================] - 2s 2ms/step - loss: 0.0629 - acc: 0.9834
938/938 [==============================] - 2s 2ms/step - loss: 0.0174 - acc: 0.9943
Epoch 1/5
94/94 [==============================] - 1s 4ms/step - loss: 0.5808 - acc: 0.8187
Epoch 2/5
94/94 [==============================] - 0s 4ms/step - loss: 0.1993 - acc: 0.9417
Epoch 3/5
94/94 [==============================] - 0s 4ms/step - loss: 0.1137 - acc: 0.9657
Epoch 4/5
94/94 [==============================] - 0s 3ms/step - loss: 0.0763 - acc: 0.9767
Epoch 5/5
94/94 [==============================] - 0s 4ms/step - loss: 0.0448 - acc: 0.9863
938/938 [==============================] - 2s 2ms/step - loss: 0.1672 - acc: 0.9527
94/94 [==============================] - 0s 2ms/step - loss: 0.0271 - acc: 0.9920
Epoch 1/5
305/305 [==============================] - 1s 3ms/step - loss: 0.3221 - acc: 0.8996
Epoch 2/5
305/305 [==============================] - 1s 3ms/step - loss: 0.1005 - acc: 0.9695
Epoch 3/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0605 - acc: 0.9819
Epoch 4/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0376 - acc: 0.9886
Epoch 5/5
305/305 [==============================] - 1s 3ms/step - loss: 0.0277 - acc: 0.9916
938/938 [==============================] - 2s 2ms/step - loss: 0.0866 - acc: 0.9778
305/305 [==============================] - 1s 2ms/step - loss: 0.0104 - acc: 0.9962
Epoch 1/5
516/516 [==============================] - 2s 3ms/step - loss: 0.2285 - acc: 0.9305
Epoch 2/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0712 - acc: 0.9782
Epoch 3/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0421 - acc: 0.9876
Epoch 4/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0270 - acc: 0.9915
Epoch 5/5
516/516 [==============================] - 2s 3ms/step - loss: 0.0199 - acc: 0.9945
938/938 [==============================] - 2s 2ms/step - loss: 0.0706 - acc: 0.9822
516/516 [==============================] - 1s 2ms/step - loss: 0.0093 - acc: 0.9972
Epoch 1/5
727/727 [==============================] - 3s 3ms/step - loss: 0.1980 - acc: 0.9400
Epoch 2/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0616 - acc: 0.9808
Epoch 3/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0386 - acc: 0.9879
Epoch 4/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0257 - acc: 0.9923
Epoch 5/5
727/727 [==============================] - 2s 3ms/step - loss: 0.0174 - acc: 0.9949
938/938 [==============================] - 2s 2ms/step - loss: 0.0631 - acc: 0.9852
727/727 [==============================] - 1s 2ms/step - loss: 0.0088 - acc: 0.9977
Epoch 1/5
938/938 [==============================] - 3s 3ms/step - loss: 0.1621 - acc: 0.9503
Epoch 2/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0535 - acc: 0.9842
Epoch 3/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0337 - acc: 0.9898
Epoch 4/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0237 - acc: 0.9925
Epoch 5/5
938/938 [==============================] - 3s 3ms/step - loss: 0.0181 - acc: 0.9944
938/938 [==============================] - 2s 2ms/step - loss: 0.0538 - acc: 0.9859
938/938 [==============================] - 2s 2ms/step - loss: 0.0088 - acc: 0.9976
train_score
array([[0.99599999, 0.99199998],
[0.99764103, 0.99620515],
[0.99757576, 0.99715149],
[0.99582797, 0.99772042],
[0.99426669, 0.99756664]])
sklearn_evaluation.plot.learning_curve(train_score, test_score, train_size)
<AxesSubplot:title={'center':'Learning Curve'}, xlabel='Training examples', ylabel='Score mean'>
Lasso > L1
Ridge > L2
ElasticNet
CNN 도 linear 기 때문에 L1, L2 사용할 수 있다.
과적합(Overfitting)을 막는 방법
- train data 가 많은 것(확실하다)
- Regularizer 사용
- Drop out : catastrophic forgetting 을 줄여주는 역할도 함, ensemble 효과있음
- Batch Normalizer 사용
- Model Checkpoint / Early Stopping
tensorflow overfitting
- tf.keras.regularizers.L1L2
- tf.keras.regularizers.l1_l2
- tf.keras.layers.Dropout
- tf.keras.callbacks.EarlyStopping
- tf.keras.callbacks.LearningRateScheduler
- tf.keras.callbacks.ReduceLROnPlateau
