is1394 · September 1, 2017 02:56 · Feb 21, 2017 · Feb 21, 2017 · Feb 20, 2017 · Feb 20, 2017
diff --git a/xor_example2.py b/xor_example2.py
@@ -13,7 +13,7 @@
 x = np.array([[0, 0], [0, 1], [1, 0], [1, 1]], dtype=np.float32)
 y1 = np.logical_xor(x[:, 0], x[:, 1])
 y0 = np.logical_not(y1)
-y = np.reshape([y0, y1], (-1, 2)).astype(np.float32)
+y = np.array([y0, y1]).T.astype(np.float32)
 
 # train
 model.fit(x, y, nb_epoch=15000)

diff --git a/mnist_example.py b/mnist_example.py
@@ -23,7 +23,7 @@
 model.add(Activation("softmax"))
 
 model.compile(
-    optimizer="sgd", loss="categorical_crossentropy", metrics=['accuracy'])
+    optimizer="sgd", loss="categorical_crossentropy", metrics=["accuracy"])
 print(model.to_yaml()) # show model
 
 

diff --git a/xor_example2.py b/xor_example2.py
@@ -0,0 +1,32 @@
+from keras.models import Sequential
+from keras.layers import Activation, Dense
+
+# xor model
+model = Sequential()
+model.add(Dense(4, input_dim=2, activation="relu"))
+model.add(Dense(2, activation="softmax"))
+model.compile(optimizer="sgd", loss="categorical_crossentropy")
+
+# training data
+import numpy as np
+
+x = np.array([[0, 0], [0, 1], [1, 0], [1, 1]], dtype=np.float32)
+y1 = np.logical_xor(x[:, 0], x[:, 1])
+y0 = np.logical_not(y1)
+y = np.reshape([y0, y1], (-1, 2)).astype(np.float32)
+
+# train
+model.fit(x, y, nb_epoch=15000)
+
+print("\n[weight and bias]")
+for layer in model.layers:
+    print(layer.get_weights())
+    pass
+
+
+# predict
+pred_x = x[0:4]
+pred_y = model.predict(pred_x)
+
+print("\n[predicate]")
+print([pred_x, pred_y]) # => [[1, 0], [0, 1], [0, 1], [1, 0]]?
diff --git a/xor_example.py b/xor_example.py
@@ -3,22 +3,28 @@
 
 # xor model
 model = Sequential()
-model.add(Dense(2, input_dim=2))
-model.add(Activation("tanh"))
-model.add(Dense(1))
-model.add(Activation("linear"))
+model.add(Dense(2, input_dim=2, activation="tanh"))
+model.add(Dense(1, activation="linear"))
 model.compile(optimizer="sgd", loss="mse")
 
-# data
+# training data
 import numpy as np
 
-x = np.array([[0, 0], [0, 1], [1, 0], [1, 1]] * 10000, dtype=np.float32)
+x = np.array([[0, 0], [0, 1], [1, 0], [1, 1]], dtype=np.float32)
 y = np.logical_xor(x[:, 0], x[:, 1]).astype(np.float32).reshape((-1, 1))
 
-
 # train
-model.fit(x, y)
+model.fit(x, y, nb_epoch=15000)
+
+print("\n[weight and bias]")
+for layer in model.layers:
+    print(layer.get_weights())
+    pass
+
 
 # predict
-pred_y = model.predict(x[0:4])
-print(pred_y)
+pred_x = x[0:4]
+pred_y = model.predict(pred_x)
+
+print("\n[predicate]")
+print([pred_x, pred_y]) # => [0, 1, 1, 0] ?
diff --git a/xor_example.py b/xor_example.py
@@ -0,0 +1,24 @@
+from keras.models import Sequential
+from keras.layers import Activation, Dense
+
+# xor model
+model = Sequential()
+model.add(Dense(2, input_dim=2))
+model.add(Activation("tanh"))
+model.add(Dense(1))
+model.add(Activation("linear"))
+model.compile(optimizer="sgd", loss="mse")
+
+# data
+import numpy as np
+
+x = np.array([[0, 0], [0, 1], [1, 0], [1, 1]] * 10000, dtype=np.float32)
+y = np.logical_xor(x[:, 0], x[:, 1]).astype(np.float32).reshape((-1, 1))
+
+
+# train
+model.fit(x, y)
+
+# predict
+pred_y = model.predict(x[0:4])
+print(pred_y)
diff --git a/curve_example.py b/curve_example.py
@@ -0,0 +1,40 @@
+# simple LSTM model
+from keras.models import Sequential
+from keras.layers import Activation, Dense, LSTM
+
+model = Sequential()
+model.add(LSTM(128, input_shape=(32, 1))) # 32 timespan 1 value unit
+model.add(Dense(1))
+model.add(Activation("linear"))
+model.compile(optimizer="rmsprop", loss="mse")
+
+# several sin curve data for train
+import numpy as np
+
+for cycle in range(4, 36):
+    data = np.sin(np.arange(360) * (np.pi / cycle))
+    # chunk to 32-timespan and its 1-next value
+    x_train = np.reshape(
+        [data[i:i+32] for i in range(len(data) - 32 - 1)], (-1, 32, 1))
+    y_train = np.reshape(
+        [data[i+32] for i in range(len(data) - 32 - 1)], (-1, 1))
+    model.fit(x_train, y_train, verbose = False)
+    pass
+
+# make initial curve data
+x = np.sin(np.arange(32) * np.pi / 18  + 3 * np.pi / 4)
+# generate subsequent curve
+for i in range(300):
+    y = model.predict(x[-32:].reshape(1, 32, 1))
+    x = np.append(x, y)
+    pass
+
+# plot curve
+import matplotlib.pyplot as plt
+
+fig = plt.figure()
+sub = fig.add_subplot(1, 1, 1)
+sub.plot(x[32:])
+fig.show()
+
+input("quit to enter> ")
diff --git a/mnist_example.py b/mnist_example.py
@@ -4,9 +4,9 @@
 from keras.layers.convolutional import Convolution2D, MaxPooling2D
 
 model = Sequential()
-# count of convolution filter=32, row of filter=3, col of filter=3, 
+# count of convolution filters=32, rows of a filter=3, cols of a filter=3, 
 # note: input_shape format is different by the backend: tensorflow or theano
-#       tensorflow: (rows, cols, channels)  # number of color channels
+#       tensorflow: (rows, cols, channels)  # count of color channels
 model.add(Convolution2D(32, 3, 3, input_shape=(28, 28, 1)))
 model.add(Activation("relu"))
 model.add(Convolution2D(64, 3, 3))

diff --git a/mnist_example.py b/mnist_example.py
@@ -53,7 +53,7 @@
 
 
 # [predict by single data]
-result = model.predict(x_test[0:1]) # numpy array of a single data
+result = model.predict(x_test[0:1]) # single element numpy array of a data
 print("[result] {} may be {}".format(result[0], result[0].argmax()))
 print("[answer] {} as {}".format(y_test[0], y_test_raw[0]))
 

diff --git a/mnist_example.py b/mnist_example.py
@@ -53,7 +53,7 @@
 
 
 # [predict by single data]
-result = model.predict(x_test[0:1]) # array of a single data
+result = model.predict(x_test[0:1]) # numpy array of a single data
 print("[result] {} may be {}".format(result[0], result[0].argmax()))
 print("[answer] {} as {}".format(y_test[0], y_test_raw[0]))
 

diff --git a/mnist_example.py b/mnist_example.py
@@ -53,7 +53,7 @@
 
 
 # [predict by single data]
-result = model.predict(x_test[0:1])
+result = model.predict(x_test[0:1]) # array of a single data
 print("[result] {} may be {}".format(result[0], result[0].argmax()))
 print("[answer] {} as {}".format(y_test[0], y_test_raw[0]))
 

diff --git a/mnist_example.py b/mnist_example.py
@@ -5,7 +5,7 @@
 
 model = Sequential()
 # count of convolution filter=32, row of filter=3, col of filter=3, 
-# note: input_shape is different by the backend: tensorflow or theano
+# note: input_shape format is different by the backend: tensorflow or theano
 #       tensorflow: (rows, cols, channels)  # number of color channels
 model.add(Convolution2D(32, 3, 3, input_shape=(28, 28, 1)))
 model.add(Activation("relu"))

diff --git a/mnist_example.py b/mnist_example.py
@@ -0,0 +1,69 @@
+# [define network model]
+from keras.models import Sequential
+from keras.layers import Activation, Dense, Dropout, Flatten
+from keras.layers.convolutional import Convolution2D, MaxPooling2D
+
+model = Sequential()
+# count of convolution filter=32, row of filter=3, col of filter=3, 
+# note: input_shape is different by the backend: tensorflow or theano
+#       tensorflow: (rows, cols, channels)  # number of color channels
+model.add(Convolution2D(32, 3, 3, input_shape=(28, 28, 1)))
+model.add(Activation("relu"))
+model.add(Convolution2D(64, 3, 3))
+model.add(Activation("relu"))
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(Dropout(0.5))
+
+model.add(Flatten())
+model.add(Dense(128))
+model.add(Activation("relu"))
+model.add(Dropout(0.5))
+
+model.add(Dense(10))
+model.add(Activation("softmax"))
+
+model.compile(
+    optimizer="sgd", loss="categorical_crossentropy", metrics=['accuracy'])
+print(model.to_yaml()) # show model
+
+
+# [prepare dataset]
+from keras.datasets import mnist
+from keras.utils import np_utils
+
+# mnist dataset as monoral image and its number
+# x: 0-255 of 28x28, y: 0-9, train: 60000 x and y, test: 10000 x and y
+(x_train_raw, y_train_raw), (x_test_raw, y_test_raw) = mnist.load_data()
+
+# convert (samples, rows, cols) to (samples, rows, cols, channels)
+x_train = x_train_raw.reshape(*x_train_raw.shape, 1).astype("float32") / 255
+x_test = x_test_raw.reshape(*x_test_raw.shape, 1).astype("float32") / 255
+# convert 0-9 data to 10 categorical array: e.g. 2 => [0.0.1,0.0, 0,0,0,0,0]
+y_train = np_utils.to_categorical(y_train_raw, 10)
+y_test = np_utils.to_categorical(y_test_raw, 10)
+
+
+# [train by train dataset] wait an hour
+history = model.fit(x_train, y_train, validation_data=(x_test, y_test))
+
+
+# [check by test dataset]
+score = model.evaluate(x_test, y_test)
+print("[score] {}, accuracy: {}".format(*score))
+
+
+# [predict by single data]
+result = model.predict(x_test[0:1])
+print("[result] {} may be {}".format(result[0], result[0].argmax()))
+print("[answer] {} as {}".format(y_test[0], y_test_raw[0]))
+
+
+#############################################################################
+# [setup keras]
+# $ python3 -m venv keras
+# $ ./keras/bin/pip install six
+# $ ./keras/bin/pip install https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-0.12.1-py3-none-any.whl
+# $ ./keras/bin/pip install keras
+# $ ./keras/bin/pip install h5py # for save and load model weight
+#
+# $ ./keras/bin/python3 mnist_example.py