def plot_tSNE(model, filename="data/RML2016.10a_dict.pkl"):
from keras.models import Model
from sklearn.manifold import TSNE
from sklearn.decomposition import PCA
(mods, snrs, lbl), (X_train, Y_train), (X_test, Y_test), (train_idx, test_idx) = \
rmldataset2016.load_data(filename, train_rate=0.80)
#设计中间层输出的模型
dense2_model = Model(inputs=model.input,
outputs=model.get_layer('dense1').output)
#提取snr下的数据进行测试
for snr in [s for s in snrs if s > 14]:
test_SNRs = [lbl[x][1] for x in test_idx] #lbl: list(mod,snr)
test_X_i = X_test[np.where(np.array(test_SNRs) == snr)]
test_Y_i = Y_test[np.where(np.array(test_SNRs) == snr)]
#计算中间层输出
dense2_output = dense2_model.predict(test_X_i, batch_size=32)
Y_true = np.argmax(test_Y_i, axis=1)
#PCA降维到50以内
pca = PCA(n_components=50)
dense2_output_pca = pca.fit_transform(dense2_output)
#t-SNE降为2
tsne = TSNE(n_components=2, perplexity=5)
Y_sne = tsne.fit_transform(dense2_output_pca)
fig = plt.figure(figsize=(14, 12))
# 散点图
# plt.scatter(Y_sne[:,0],Y_sne[:,1],s=5.,color=plt.cm.Set1(Y_true / 11.))
# 标签图
data = Y_sne
x_min, x_max = np.min(data, 0), np.max(data, 0)
data = (data - x_min) / (x_max - x_min)
for i in range(Y_sne.shape[0]):
plt.text(data[i, 0],
data[i, 1],
str(Y_true[i]),
color=plt.cm.Set1(Y_true[i] / 11.),
fontdict={
'weight': 'bold',
'size': 9
})
plt.title('t-SNR at snr:{}'.format(snr))
# plt.legend() # 显示图示
fig.show()
def predict(model, filename="data/RML2016.10a_dict.pkl"):
(mods, snrs, lbl), (X_train, Y_train), (X_test, Y_test), (train_idx, test_idx) = \
rmldataset2016.load_data(filename, train_rate=0.5)
# Plot confusion matrix
test_Y_hat = model.predict(X_test, batch_size=batch_size)
confnorm, _, _ = mltools.calculate_confusion_matrix(
Y_test, test_Y_hat, classes)
mltools.plot_confusion_matrix(confnorm, labels=classes)
# Plot confusion matrix
acc = {}
for snr in snrs:
# extract classes @ SNR
# test_SNRs = map(lambda x: lbl[x][1], test_idx)
test_SNRs = [lbl[x][1] for x in test_idx]
test_X_i = X_test[np.where(np.array(test_SNRs) == snr)]
test_Y_i = Y_test[np.where(np.array(test_SNRs) == snr)]
# estimate classes
test_Y_i_hat = model.predict(test_X_i)
confnorm_i, cor, ncor = mltools.calculate_confusion_matrix(
test_Y_i, test_Y_i_hat, classes)
acc[snr] = 1.0 * cor / (cor + ncor)
mltools.plot_confusion_matrix(
confnorm_i,
labels=classes,
title="ConvNet Confusion Matrix (SNR=%d)(ACC=%2f)" %
(snr, 100.0 * acc[snr]))
# Save results to a pickle file for plotting later
print(acc)
fd = open('predictresult/cnn2_d0.5.dat', 'wb')
pickle.dump(("CNN2", 0.5, acc), fd)
# Plot accuracy curve
plt.plot(snrs, [acc[i] for i in snrs])
plt.xlabel("Signal to Noise Ratio")
plt.ylabel("Classification Accuracy")
plt.title("CNN2 Classification Accuracy on RadioML 2016.10 Alpha")
plt.show()
def main():
import rmldataset2016
import numpy as np
(mods,snrs,lbl),(X_train,Y_train),(X_test,Y_test),(train_idx,test_idx) = \
rmldataset2016.load_data(filename ="RML2016.10a_dict.pkl", train_rate = 0.2)
one_sample = X_test[0]
print(np.shape(one_sample))
print(one_sample[0:2])
print(np.max(one_sample,axis=1))
one_sample = np.power(one_sample,2)
one_sample = np.sqrt(one_sample[0,:]+one_sample[1,:])
plt.figure()
plt.title('Training Samples')
one_sample_t = np.arange(128)
plt.plot(one_sample_t,one_sample)
# plt.scatter()
plt.grid()
plt.show()
sum_sample = np.sum(one_sample)
print(sum_sample)
def predict(model,filename="data/RML2016.10a_dict.pkl"):
(mods, snrs, lbl), (X_train, Y_train), (X_test, Y_test), (train_idx, test_idx) = \
rmldataset2016.load_data(filename, train_rate=0.5)
# Plot confusion matrix
test_Y_hat = model.predict(X_test, batch_size=batch_size)
confnorm,_,_ = mltools.calculate_confusion_matrix(Y_test,test_Y_hat,classes)
mltools.plot_confusion_matrix(confnorm, labels=classes,save_filename='figure/total_confusion')
# Plot confusion matrix
acc = {}
acc_mod_snr = np.zeros( (len(classes),len(snrs)) )
i = 0
for snr in snrs:
# extract classes @ SNR
# test_SNRs = map(lambda x: lbl[x][1], test_idx)
test_SNRs = [lbl[x][1] for x in test_idx]
test_X_i = X_test[np.where(np.array(test_SNRs) == snr)]
test_Y_i = Y_test[np.where(np.array(test_SNRs) == snr)]
# estimate classes
test_Y_i_hat = model.predict(test_X_i)
confnorm_i,cor,ncor = mltools.calculate_confusion_matrix(test_Y_i,test_Y_i_hat,classes)
acc[snr] = 1.0 * cor / (cor + ncor)
mltools.plot_confusion_matrix(confnorm_i, labels=classes, title="ConvNet Confusion Matrix (SNR=%d)(ACC=%2f)" % (snr,100.0*acc[snr]),save_filename="figure/Confusion(SNR=%d)(ACC=%2f).png" % (snr,100.0*acc[snr]))
acc_mod_snr[:,i] = np.round(np.diag(confnorm_i)/np.sum(confnorm_i,axis=1),3)
i = i +1
#plot acc of each mod in one picture
dis_num=11
for g in range(int(np.ceil(acc_mod_snr.shape[0]/dis_num))):
assert (0 <= dis_num <= acc_mod_snr.shape[0])
beg_index = g*dis_num
end_index = np.min([(g+1)*dis_num,acc_mod_snr.shape[0]])
plt.figure(figsize=(12, 10))
plt.xlabel("Signal to Noise Ratio")
plt.ylabel("Classification Accuracy")
plt.title("Classification Accuracy for Each Mod")
for i in range(beg_index,end_index):
plt.plot(snrs, acc_mod_snr[i], label=classes[i])
# 设置数字标签
for x, y in zip(snrs, acc_mod_snr[i]):
plt.text(x, y, y, ha='center', va='bottom', fontsize=8)
plt.legend()
plt.grid()
plt.savefig('figure/acc_with_mod_{}.png'.format(g+1))
plt.close()
#save acc for mod per SNR
fd = open('predictresult/acc_for_mod_on_cldnn.dat', 'wb')
pickle.dump(('128','cldnn', acc_mod_snr), fd)
fd.close()
# Save results to a pickle file for plotting later
print(acc)
fd = open('predictresult/CLDNN_dr0.5.dat','wb')
pickle.dump( ("1D", 0.5, acc) , fd )
# Plot accuracy curve
plt.plot(snrs, list(map(lambda x: acc[x], snrs)))
plt.xlabel("Signal to Noise Ratio")
plt.ylabel("Classification Accuracy")
plt.title("CLDNN Classification Accuracy on RadioML 2016.10 Alpha")
plt.tight_layout()
plt.savefig('figure/each_acc.png')
from keras.models import model_from_json
#from keras.utils.vis_utils import plot_model
import mltools,rmldataset2016
#import rmlmodels.CNN2Model as cnn2
#import rmlmodels.ResNetLikeModel as resnet
#import rmlmodels.VGGLikeModel as vggnet
import rmlmodels.CLDNNLikeModel as cldnn
#set Keras data format as channels_first
K.set_image_data_format('channels_last')
print(K.image_data_format())
(mods,snrs,lbl),(X_train,Y_train),(X_test,Y_test),(train_idx,test_idx) = \
rmldataset2016.load_data(filename ="data/RML2016.10a_dict.pkl", train_rate = 0.5)
in_shp = list(X_train.shape[1:])
print(X_train.shape, in_shp)
classes = mods
print(classes)
# Build VT-CNN2 Neural Net model using Keras primitives --
# - Reshape [N,2,128] to [N,2,128,1] on input
# - Pass through 2 2DConv/ReLu layers
# - Pass through 2 Dense layers (ReLu and Softmax)
# - Perform categorical cross entropy optimization
# Set up some params
nb_epoch = 100 # number of epochs to train on
batch_size = 1024 # training batch size
