Created
December 17, 2017 10:36
-
-
Save adiamb/fec35bc8e2a1ec59639abcdb0696bceb to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import pandas as pd | |
import numpy as np | |
import keras | |
from sklearn.model_selection import train_test_split | |
from sklearn.preprocessing import OneHotEncoder | |
from sklearn.preprocessing import StandardScaler | |
import matplotlib.pyplot as plt | |
%matplotlib inline | |
from sklearn.model_selection import train_test_split | |
from sklearn.metrics import confusion_matrix | |
from keras.utils.np_utils import to_categorical # convert to one-hot-encoding | |
from keras.models import Sequential | |
from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D, BatchNormalization | |
from keras.optimizers import Adam | |
from keras.preprocessing.image import ImageDataGenerator | |
from keras.callbacks import LearningRateScheduler | |
import numpy | |
# fix random seed for reproducibility | |
numpy.random.seed(7) | |
#from sklearn.prep | |
### read in the training data | |
mnst_ = pd.read_csv('train_MINST.csv', header=0) | |
mnst_test= pd.read_csv('test_MINST.csv', header=0) | |
one_hot_labels = keras.utils.to_categorical(mnst_.label.values, num_classes=10) | |
train_n = mnst_.shape[0] | |
test_n= mnst_test.shape[0] | |
## concat the train and tes frames | |
total=pd.concat([mnst_.iloc[:, 1:], mnst_test], axis=0, ignore_index=True) | |
### scale the values | |
std_sc=StandardScaler() | |
std_sc.fit(total.values) | |
total_sc=std_sc.transform(total.values) | |
### build a tf network | |
train_x, test_x, train_y, test_y = train_test_split(total_sc[:train_n], one_hot_labels, | |
test_size=0.33, random_state=42) | |
model = Sequential() | |
model.add(Dense(300, activation='sigmoid', input_dim=train_x.shape[1])) | |
model.add(Dropout(0.2)) | |
model.add(Dense(150, activation='sigmoid')) | |
model.add(Dropout(0.2)) | |
model.add(Dense(10, activation='softmax')) | |
model.compile(optimizer='rmsprop', | |
loss='categorical_crossentropy', | |
metrics=['accuracy']) | |
his_=model.fit(train_x, train_y, | |
epochs=20, batch_size=32, | |
validation_data=(test_x,test_y), shuffle=True) | |
test_ = total_sc[train_n:] | |
pre_=model.predict_classes(test_) | |
out=pd.DataFrame(index=range(1, len(pre_)+1), columns=['ImageID','Label']) | |
out.loc[:, 'ImageID'] = range(1, len(pre_)+1) | |
out.loc[:, 'Label'] = pre_.tolist() | |
out.to_csv('/Users/adityaambati/Desktop/sub3_minst.csv', index=False) | |
#train_x, test_x, train_y, test_y = train_test_split(mnst_.iloc[:, 1:].values, one_hot_labels,test_size=0.33, random_state=42) | |
fig, ax = plt.subplots(2, 1, figsize=(12,6)) | |
ax[0].plot(train_x[0]) | |
ax[0].set_title('784x1 data') | |
ax[1].imshow(train_x[0].reshape(28,28), cmap='gray') | |
ax[1].set_title('28x28 data') | |
train_cnn=train_x.reshape(-1, 28, 28, 1) | |
test_cnn=test_x.reshape(-1, 28, 28, 1) | |
test_cnn_f=total_sc[train_n:].reshape(-1, 28, 28, 1) | |
#train_cnn=train_cnn.astype('float32')/255 | |
#test_cnn=test_cnn.astype('float32')/255 | |
#test_cnn_f=test_cnn_f.astype('float32')/255 | |
model = Sequential() | |
model.add(Conv2D(filters = 16, (3, 3), activation='relu', | |
input_shape = (28, 28, 1), padding='same')) | |
model.add(BatchNormalization()) | |
model.add(Conv2D(filters = 16, (3, 3), activation='relu')) | |
model.add(BatchNormalization()) | |
#model.add(Conv2D(filters = 16, kernel_size = (3, 3), activation='relu')) | |
#model.add(BatchNormalization()) | |
model.add(MaxPool2D(strides=(2,2))) | |
model.add(Dropout(0.25)) | |
model.add(Conv2D(filters = 32, (3, 3), activation='relu', padding='same')) | |
model.add(BatchNormalization()) | |
model.add(Conv2D(filters = 32, (3, 3), activation='relu')) | |
model.add(BatchNormalization()) | |
#model.add(Conv2D(filters = 32, kernel_size = (3, 3), activation='relu')) | |
#model.add(BatchNormalization()) | |
model.add(MaxPool2D(strides=(2,2))) | |
model.add(Dropout(0.25)) | |
model.add(Flatten()) | |
model.add(Dense(512, activation='relu')) | |
model.add(Dropout(0.25)) | |
model.add(Dense(1024, activation='relu')) | |
model.add(Dropout(0.5)) | |
model.add(Dense(10, activation='softmax')) | |
model.compile(loss='categorical_crossentropy', optimizer = Adam(lr=1e-4), metrics=["accuracy"]) | |
datagen = ImageDataGenerator(zoom_range = 0.1, | |
height_shift_range = 0.1, | |
width_shift_range = 0.1, | |
rotation_range = 10) | |
annealer = LearningRateScheduler(lambda x: 1e-3 * 0.9 ** x) | |
hist = model.fit_generator(datagen.flow(train_cnn, train_y, batch_size=16), | |
steps_per_epoch=500, | |
epochs=100, | |
callbacks=[annealer], | |
validation_data=(test_cnn, test_y)) | |
pre_=model.predict_classes(test_cnn_f) | |
out=pd.DataFrame(index=range(1, len(pre_)+1), columns=['ImageID','Label']) | |
out.loc[:, 'ImageID'] = range(1, len(pre_)+1) | |
out.loc[:, 'Label'] = pre_.tolist() | |
out.to_csv('/Users/adityaambati/Desktop/sub3_cnn_minst.csv', index=False) | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Thank you for sharing this nice and clean code of yours. It is really helpful. I borrowed your learning rate annealing scheme while training a 50 layer ResNET on whale fluke images. But when I monitor LR at each epoch I see no change! How exactly does your function lambda x: 1e-3 * 0.9 ** x behave? Is it something like f(eta) = (0.9^eta)/1000? If is that so the LR should quickly converge to zero after a few epochs. I'm a bit confused, here. Any help is appreciated.