From dd6fd99d20c86ab0790616bfee9d0c1a16760d09 Mon Sep 17 00:00:00 2001
From: orbxball <orbxball@protonmail.com>
Date: Fri, 30 Jun 2017 01:31:56 +0800
Subject: [PATCH] hw3: ensemble

---
 hw3/ensemble1.py | 124 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 124 insertions(+)
 create mode 100644 hw3/ensemble1.py

diff --git a/hw3/ensemble1.py b/hw3/ensemble1.py
new file mode 100644
index 0000000..32706f9
--- /dev/null
+++ b/hw3/ensemble1.py
@@ -0,0 +1,124 @@
+#!/usr/bin/env python
+import sys, os
+import numpy as np
+from keras.utils import to_categorical
+from keras.preprocessing.image import ImageDataGenerator
+from keras.models import Sequential
+from keras.layers import Dense, Activation, Flatten, Dropout, LeakyReLU
+from keras.layers import Conv2D, MaxPooling2D, AveragePooling2D
+from keras.layers import BatchNormalization
+from keras import losses
+from keras import optimizers
+from keras.callbacks import CSVLogger, ModelCheckpoint
+
+# Parameter
+height = width = 48
+num_classes = 7
+input_shape = (height, width, 1)
+batch_size = 128
+epochs = 50
+zoom_range = 0.2
+model_name = 'en1.h5'
+isValid = 1
+
+# Read the train data
+with open(sys.argv[1], "r+") as f:
+  line = f.read().strip().replace(',', ' ').split('\n')[1:]
+  raw_data = ' '.join(line)
+  length = width*height+1 #1 is for label
+  data = np.array(raw_data.split()).astype('float').reshape(-1, length)
+  X = data[:, 1:]
+  Y = data[:, 0]
+  X /= 255
+  Y = Y.reshape(Y.shape[0], 1)
+  Y = to_categorical(Y, num_classes)
+
+# Change data into CNN format
+X = X.reshape(X.shape[0], height, width, 1)
+
+# Split the data
+if isValid:
+  valid_num = 3000
+  X_train, Y_train = X[:-valid_num], Y[:-valid_num]
+  X_valid, Y_valid = X[-valid_num:], Y[-valid_num:]
+
+else:
+  X_train, Y_train = X, Y
+
+# Construct the model
+model = Sequential()
+model.add(Conv2D(32, (3, 3), padding='same', input_shape=input_shape))
+model.add(LeakyReLU(alpha=0.03))
+model.add(BatchNormalization())
+model.add(Conv2D(32, (3, 3), padding='same'))
+model.add(LeakyReLU(alpha=0.03))
+model.add(BatchNormalization())
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(Dropout(0.3))
+
+model.add(Conv2D(128, (3, 3), padding='same'))
+model.add(LeakyReLU(alpha=0.03))
+model.add(BatchNormalization())
+model.add(Conv2D(128, (3, 3), padding='same'))
+model.add(LeakyReLU(alpha=0.03))
+model.add(BatchNormalization())
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(Dropout(0.4))
+
+model.add(Conv2D(512, (3, 3), padding='same'))
+model.add(LeakyReLU(alpha=0.03))
+model.add(BatchNormalization())
+model.add(Conv2D(512, (3, 3), padding='same'))
+model.add(LeakyReLU(alpha=0.03))
+model.add(BatchNormalization())
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(Dropout(0.5))
+
+model.add(Flatten())
+
+model.add(Dense(128, activation='relu'))
+model.add(BatchNormalization())
+model.add(Dropout(0.5))
+model.add(Dense(num_classes))
+model.add(Activation('softmax'))
+
+model.summary()
+
+# Compile the model
+model.compile(loss='categorical_crossentropy',
+              optimizer='adam',
+              metrics=['accuracy'])
+
+# Image PreProcessing
+train_gen = ImageDataGenerator(rotation_range=25,
+                              width_shift_range=0.1,
+                              height_shift_range=0.1,
+                              shear_range=0.1,
+                              zoom_range=[1-zoom_range, 1+zoom_range],
+                              horizontal_flip=True)
+train_gen.fit(X_train)
+
+# Callbacks
+callbacks = []
+modelcheckpoint = ModelCheckpoint('en1_ckpt/weights.{epoch:03d}-{val_acc:.5f}.h5', monitor='val_acc', save_best_only=True)
+callbacks.append(modelcheckpoint)
+csv_logger = CSVLogger('en1_log.csv', separator=',', append=False)
+callbacks.append(csv_logger)
+
+
+# Fit the model
+if isValid:
+  model.fit_generator(train_gen.flow(X_train, Y_train, batch_size=batch_size),
+                    steps_per_epoch=10*X_train.shape[0]//batch_size,
+                    epochs=epochs,
+                    callbacks=callbacks,
+                    validation_data=(X_valid, Y_valid))
+else:
+  model.fit_generator(train_gen.flow(X_train, Y_train, batch_size=batch_size),
+                    steps_per_epoch=10*X_train.shape[0]//batch_size,
+                    epochs=epochs,
+                    callbacks=callbacks)
+
+
+# Save model
+model.save(model_name)