keras兼容了tensorflow，也兼容了Theano
https://www.tensorflow.org/tutorials/
Basic
import tensorflow as tf import numpy as np import pandas as pd from tensorflow import keras from tensorflow.keras import layers tf.keras.backend.clear_session() COLUMN_NAMES = ['SepalLength', 'SepalWidth', 'PetalLength', 'PetalWidth', 'Species'] SPECIES = ['Setosa', 'Versicolor', 'Virginica'] # Get Train and Test Data Set Directly from Tensorflow so we won't do train and test spilt in this case #从tensorflow的网址上将数据集下载到本地 train_path = tf.keras.utils.get_file( "iris_training.csv", "https://storage.googleapis.com/download.tensorflow.org/data/iris_training.csv") test_path = tf.keras.utils.get_file( "iris_test.csv", "https://storage.googleapis.com/download.tensorflow.org/data/iris_test.csv") train = pd.read_csv(train_path, names=COLUMN_NAMES, header=0) test = pd.read_csv(test_path, names=COLUMN_NAMES, header=0) print(train.head()) # Seperate Features and Labels by removing Species from train and test set and storing it in train_Y and test_y train_y = train.pop('Species') test_y = test.pop('Species') #---------------------- #Create Model #---------------------- inputs = keras.Input(shape=(4,), name='flowers') #inputs是tensor类型的 Tensor("flowers:0", shape=(None, 4), dtype=float32) x = layers.Dense(units=64, activation='relu', name='layer_1')(inputs) x = layers.Dense(units=64, activation='relu', name='layer_2')(x) outputs = layers.Dense(units=3, activation='softmax', name='predictions')(x) model = keras.Model(inputs=inputs, outputs=outputs) tensorboard_callback = keras.callbacks.TensorBoard(log_dir="logs") # Optimizer which helps minimize the loss function and it is a compulsory argument model.compile(optimizer=keras.optimizers.RMSprop(), # Loss function to minimize (Also a compulsory argument) loss=keras.losses.SparseCategoricalCrossentropy(), # List of metrics to monitor metrics=[keras.metrics.SparseCategoricalAccuracy()]) #---------------------- #Train #---------------------- X = train.values[:100] y = train_y.values[:100] x_val = train.values[100:] y_val = train_y.values[100:] history = model.fit(X, y, batch_size=64, epochs=100, callbacks=[tensorboard_callback], # We pass some validation for # monitoring validation loss and metrics # at the end of each epoch validation_data=(x_val, y_val)) print(history.history) #---------------------- #Analyse #---------------------- print(model.summary()) keras.utils.plot_model(model, show_shapes=True) #---------------------- #Predict #---------------------- x_test = test.values y_test = test_y.values results = model.evaluate(x_test, y_test, batch_size=64) print('test loss, test acc:', results) res = model.evaluate(train.values, train_y.values, batch_size=64) print('train loss, train acc:', res) #---------------------- #Visulize #---------------------- # %tensorboard --logdir ./model/logshttps://colab.research.google.com/drive/1c7YkLvrXxwYU7yDZCEtEYGeVkILkaHt_
Examples for beginners
import tensorflow as tf tf.keras.backend.clear_session() mnist = tf.keras.datasets.mnist (x_train, y_train), (x_test, y_test) = mnist.load_data() x_train, x_test = x_train / 255.0, x_test / 255.0 model = tf.keras.models.Sequential([ tf.keras.layers.Flatten(input_shape=(28, 28)), tf.keras.layers.Dense(128, activation='relu'), tf.keras.layers.Dropout(0.2), tf.keras.layers.Dense(10, activation='softmax') ]) model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) model.fit(x_train, y_train, epochs=5) model.evaluate(x_test, y_test, verbose=2)
这个照片分类器的准确度已经达到 98%
Tensorflow Estimator
https://colab.research.google.com/drive/1wgk2mYRZLVlpVyipuJA6eXi06ZTBTxF1#scrollTo=y1jxA2nmrpbR
import tensorflow as tf import numpy as np import pandas as pd COLUMN_NAMES = ['SepalLength', 'SepalWidth', 'PetalLength', 'PetalWidth', 'Species'] SPECIES = ['Setosa', 'Versicolor', 'Virginica'] train_path = tf.keras.utils.get_file( "iris_training.csv", "https://storage.googleapis.com/download.tensorflow.org/data/iris_training.csv") test_path = tf.keras.utils.get_file( "iris_test.csv", "https://storage.googleapis.com/download.tensorflow.org/data/iris_test.csv") train = pd.read_csv(train_path, names=COLUMN_NAMES, header=0) test = pd.read_csv(test_path, names=COLUMN_NAMES, header=0) train_y = train.pop('Species') test_y = test.pop('Species') # Create Model # Input_fn function helps convert the pandas dataframe into a tensorflow dataset which is the only acceptable input to the Tensorflow Estimator while training and predicting / evaluating. # If the data is being used for training the function returns randomly shuffled training samples from the dataset in batches of 256 def input_fn(features, labels, training=True, batch_size=256): """An input function for training or evaluating""" # Convert the inputs to a Tensorflow Dataset. # print(dict(features)) dataset = tf.data.Dataset.from_tensor_slices((dict(features), labels)) # Shuffle and repeat if you are in training mode. if training: dataset = dataset.shuffle(1000).repeat() return dataset.batch(batch_size) feature_columns = [] for key in train.keys(): feature_columns.append(tf.feature_column.numeric_column(key=key)) print(feature_columns) logistic_regression_classifier = tf.estimator.LinearClassifier( feature_columns= feature_columns, # The model's location in your computer model_dir="./model/logs/LR", # The model must choose between 3 classes. n_classes=3 ) # This exercise also uses a DNNClassifier estimator which is a pre-made estimator (essentially a Multilayer Perceptron) in Tensorflow 2.0 and is designed for multi-class classification and it is the most suitable for us in Iris Dataset with 3 class labels # The current DNN Classifier is a neural network with 2 hidden layers having 60 neurons and 30 neurons each, so the model will have an input layer consisting of the 4 features, the 2 hidden layers of neurons and the output layer consisting a probability estimate for each class, hence the output will give probabilities for each of the 3 class labels. For eg: One output can be {'Setosa': 90%, 'Versicolor': 2%, 'Virginica': 8%} #Builds a DNN with 2 hidden layers with 60 and 30 hidden nodes each. DNNClassifier = tf.estimator.DNNClassifier( feature_columns= feature_columns, hidden_units=[60, 30], # The model's location in your computer model_dir="./model/logs/DNN", # The model must choose between 3 classes. n_classes=3) # Training # We train both the models for 7000 steps to compare the performance. A step is one feed forward propogation of a batch of the data through the model during training. # Note that Logistic Regression is not a neural network (Just uses one sigmoid activation function, can be though of as 1 neuron network). # However DNN Classifier is nerual network based model and it is a multilayer perceptron. logistic_regression_classifier.train( input_fn=lambda: input_fn(train, train_y, training=True), steps=7000 ) #Train the Model DNNClassifier.train( input_fn=lambda: input_fn(train, train_y, training=True), steps=7000 ) # Evaluating # Accuracies using Logistic Regression eval_result = logistic_regression_classifier.evaluate( input_fn=lambda: input_fn(test, test_y, training=False), name='LR test') print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result)) eval_result_train = logistic_regression_classifier.evaluate( input_fn=lambda: input_fn(train, train_y, training=False), name='LR train') print('\nTrain set accuracy: {accuracy:0.3f}\n'.format(**eval_result_train)) # Accuracies using MultiLayer Perceptron (DNN Classifier) # The name parameter in evaluate property of the classifier is used during visualisation of accuracy in tensorboard eval_result = DNNClassifier.evaluate( input_fn=lambda: input_fn(test, test_y, training=False), name='DNN test') print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result)) eval_result_train = DNNClassifier.evaluate( input_fn=lambda: input_fn(train, train_y, training=False), name='DNN train') print('\nTrain set accuracy: {accuracy:0.3f}\n'.format(**eval_result_train)) # Predicting # Sample Set of 3 Features and expected labels to get the predictions for expected = ['Setosa', 'Versicolor', 'Virginica'] predict_x = { 'SepalLength': [5.1, 5.9, 6.9], 'SepalWidth': [3.3, 3.0, 3.1], 'PetalLength': [1.7, 4.2, 5.4], 'PetalWidth': [0.5, 1.5, 2.1], } def input_fn_test(features, batch_size=256): # Convert the inputs to a Tensorflow Dataset without labels for using in the model to predict the class labels return tf.data.Dataset.from_tensor_slices(dict(features)).batch(batch_size) DNNPredictions = DNNClassifier.predict( input_fn=lambda: input_fn_test(predict_x)) LRPredictions = logistic_regression_classifier.predict( input_fn=lambda: input_fn_test(predict_x)) # Predictions using the Logistic Regression Model along with the probabilities for pred_dict, expec in zip(LRPredictions, expected): class_id = pred_dict['class_ids'][0] probability = pred_dict['probabilities'][class_id] print('Prediction is "{}" ({:.1f}%), expected "{}"'.format( SPECIES[class_id], 100 * probability, expec)) # Predictions using the DNN Classifier Model along with the probabilities for pred_dict, expec in zip(DNNPredictions, expected): class_id = pred_dict['class_ids'][0] probability = pred_dict['probabilities'][class_id] print('Prediction is "{}" ({:.1f}%), expected "{}"'.format( SPECIES[class_id], 100 * probability, expec))CNN
RNN
Tensorflow2.0 & Keras
猜你喜欢
转载自blog.csdn.net/hxxjxw/article/details/113408293
今日推荐
周排行