In [ ]:
CNN classifier for the planesnet data¶
In [ ]:
## import packages
import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import MinMaxScaler
from keras.models import Sequential, Model
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dense, Activation, Dropout, Flatten
from tensorflow.keras.layers import BatchNormalization
from tensorflow.keras.optimizers import Adam
from keras.callbacks import LearningRateScheduler
from keras.regularizers import l2, l1
import math
In [ ]:
## Mount drive folder
from google.colab import drive
drive.mount('/content/drive')
Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).
In [ ]:
## Read data in json format
data = pd.read_json('/content/drive/MyDrive/CommonFiles/MUSA650-Data/planesnet.json')
In [ ]:
## View the dataframe
data.head()
Out[ ]:
| data | labels | locations | scene_ids | |
|---|---|---|---|---|
| 0 | [206, 195, 187, 183, 177, 175, 174, 193, 198, ... | 1 | [-118.40497658522878, 33.940618514147936] | 20170620_175442_0e30 |
| 1 | [215, 209, 200, 196, 192, 197, 205, 168, 155, ... | 1 | [-122.392469714, 37.6176425378] | 20161212_180859_0e30 |
| 2 | [204, 214, 220, 219, 213, 205, 198, 193, 199, ... | 1 | [-122.397578597, 37.6209247852] | 20170524_181349_0e2f |
| 3 | [179, 174, 179, 178, 173, 170, 168, 168, 168, ... | 1 | [-122.214849831, 37.7203378331] | 20161110_180707_0e1f |
| 4 | [222, 222, 218, 214, 208, 205, 207, 206, 206, ... | 1 | [-117.862173435, 33.6796854072] | 20160813_184932_0c64 |
In [ ]:
## Read image data into a 2D matrix
X = np.array(data.data.tolist())
X.shape
Out[ ]:
(32000, 1200)
In [ ]:
## Reshape the image data to 20x20 image patches with 3 channels (RGB)
X = X.reshape(32000, 3, 20, 20)
X.shape
Out[ ]:
(32000, 3, 20, 20)
In [ ]:
## Standard format for RGB is "the channels at the end"; move the axis for channels to the end
X = np.moveaxis(X, 1, 3)
X.shape
Out[ ]:
(32000, 20, 20, 3)
In [ ]:
## View few images
tmpI = X[46,:,:,:]
plt.imshow(tmpI)
plt.show()
In [ ]:
## Read the labels
y = np.array(data['labels'])
y.shape
Out[ ]:
(32000,)
In [ ]:
## Split the data into training ans testing sets
X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size = 0.20)
## Scale the data
scalar = MinMaxScaler()
scalar.fit(X_tr.reshape(X_tr.shape[0], -1))
X_tr = scalar.transform(X_tr.reshape(X_tr.shape[0], -1)).reshape(X_tr.shape)
X_te = scalar.transform(X_te.reshape(X_te.shape[0], -1)).reshape(X_te.shape)
In [ ]:
## Show data size
print(X_tr.shape)
print(y_tr.shape)
print(X_te.shape)
print(y_te.shape)
(25600, 20, 20, 3) (25600,) (6400, 20, 20, 3) (6400,)
In [ ]:
## Get the shape of a single sample
dshape = X_tr.shape[1:]
dshape
Out[ ]:
(20, 20, 3)
In [ ]:
## Construct the model
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3), strides=(1,1), input_shape=dshape))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D((2,2)))
model.add(Conv2D(32, kernel_size=(3, 3), strides=(1,1)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(MaxPooling2D((2,2)))
model.add(Flatten())
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))
print(model.summary())
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d (Conv2D) (None, 18, 18, 32) 896
batch_normalization (BatchN (None, 18, 18, 32) 128
ormalization)
activation (Activation) (None, 18, 18, 32) 0
max_pooling2d (MaxPooling2D (None, 9, 9, 32) 0
)
conv2d_1 (Conv2D) (None, 7, 7, 32) 9248
batch_normalization_1 (Batc (None, 7, 7, 32) 128
hNormalization)
activation_1 (Activation) (None, 7, 7, 32) 0
max_pooling2d_1 (MaxPooling (None, 3, 3, 32) 0
2D)
flatten (Flatten) (None, 288) 0
dense (Dense) (None, 64) 18496
dropout (Dropout) (None, 64) 0
dense_1 (Dense) (None, 1) 65
=================================================================
Total params: 28,961
Trainable params: 28,833
Non-trainable params: 128
_________________________________________________________________
None
In [ ]:
## Compile the model
OPT = Adam(learning_rate=0.0001)
NUM_EPOCH = 15
B_SIZE = 64
model.compile(loss='binary_crossentropy', optimizer=OPT, metrics=['accuracy'])
In [ ]:
## Train the model
mdl_tr = model.fit(X_tr, y_tr, batch_size=B_SIZE, epochs=NUM_EPOCH, shuffle=True, validation_data=(X_te, y_te))
Epoch 1/15 400/400 [==============================] - 19s 44ms/step - loss: 0.3205 - accuracy: 0.8585 - val_loss: 0.7776 - val_accuracy: 0.7542 Epoch 2/15 400/400 [==============================] - 16s 40ms/step - loss: 0.1881 - accuracy: 0.9271 - val_loss: 0.1466 - val_accuracy: 0.9458 Epoch 3/15 400/400 [==============================] - 16s 40ms/step - loss: 0.1545 - accuracy: 0.9420 - val_loss: 0.1438 - val_accuracy: 0.9456 Epoch 4/15 400/400 [==============================] - 21s 52ms/step - loss: 0.1347 - accuracy: 0.9498 - val_loss: 0.1248 - val_accuracy: 0.9536 Epoch 5/15 400/400 [==============================] - 21s 53ms/step - loss: 0.1224 - accuracy: 0.9544 - val_loss: 0.1179 - val_accuracy: 0.9592 Epoch 6/15 400/400 [==============================] - 16s 40ms/step - loss: 0.1120 - accuracy: 0.9591 - val_loss: 0.1039 - val_accuracy: 0.9628 Epoch 7/15 400/400 [==============================] - 17s 41ms/step - loss: 0.1031 - accuracy: 0.9619 - val_loss: 0.1323 - val_accuracy: 0.9503 Epoch 8/15 400/400 [==============================] - 17s 42ms/step - loss: 0.0972 - accuracy: 0.9639 - val_loss: 0.0926 - val_accuracy: 0.9638 Epoch 9/15 400/400 [==============================] - 16s 41ms/step - loss: 0.0923 - accuracy: 0.9664 - val_loss: 0.1069 - val_accuracy: 0.9609 Epoch 10/15 400/400 [==============================] - 17s 42ms/step - loss: 0.0848 - accuracy: 0.9700 - val_loss: 0.0822 - val_accuracy: 0.9712 Epoch 11/15 400/400 [==============================] - 18s 44ms/step - loss: 0.0816 - accuracy: 0.9698 - val_loss: 0.0769 - val_accuracy: 0.9733 Epoch 12/15 400/400 [==============================] - 17s 43ms/step - loss: 0.0774 - accuracy: 0.9726 - val_loss: 0.0792 - val_accuracy: 0.9723 Epoch 13/15 400/400 [==============================] - 17s 42ms/step - loss: 0.0748 - accuracy: 0.9729 - val_loss: 0.0762 - val_accuracy: 0.9708 Epoch 14/15 400/400 [==============================] - 17s 42ms/step - loss: 0.0689 - accuracy: 0.9755 - val_loss: 0.1070 - val_accuracy: 0.9598 Epoch 15/15 400/400 [==============================] - 18s 45ms/step - loss: 0.0665 - accuracy: 0.9765 - val_loss: 0.0920 - val_accuracy: 0.9656
In [ ]:
## Plot the learning stats
fig, ax = plt.subplots(1,2, figsize=[18,6])
ax[0].plot(range(1, NUM_EPOCH+1), mdl_tr.history['loss'], c='blue', label='Training loss')
ax[0].plot(range(1, NUM_EPOCH+1), mdl_tr.history['val_loss'], c='red', label='Validation loss')
ax[0].legend()
ax[0].set_xlabel('epochs')
ax[1].plot(range(1, NUM_EPOCH+1), mdl_tr.history['accuracy'], c='blue', label='Training accuracy')
ax[1].plot(range(1, NUM_EPOCH+1), mdl_tr.history['val_accuracy'], c='red', label='Validation accuracy')
ax[1].legend()
ax[1].set_xlabel('epochs')
Out[ ]:
Text(0.5, 0, 'epochs')
