Delete network.py
Browse files- network.py +0 -422
network.py
DELETED
@@ -1,422 +0,0 @@
|
|
1 |
-
import tensorflow as tf
|
2 |
-
import os
|
3 |
-
import numpy as np
|
4 |
-
|
5 |
-
# the package needed when use Spec_Checker class
|
6 |
-
try:
|
7 |
-
from astropy.table import Table
|
8 |
-
import matplotlib.pyplot as plt
|
9 |
-
import numpy as np
|
10 |
-
import pandas as pd
|
11 |
-
import seaborn as sns
|
12 |
-
found = True
|
13 |
-
except ImportError:
|
14 |
-
found = False
|
15 |
-
print('error, pack not found####')
|
16 |
-
|
17 |
-
|
18 |
-
# https://www.tensorflow.org/guide/keras/train_and_evaluate?hl=zh-cn
|
19 |
-
class Metric_Fun(tf.keras.metrics.Metric):
|
20 |
-
"""
|
21 |
-
A customized metric.
|
22 |
-
metric = accraacy - mae.
|
23 |
-
The larger it is, the better.
|
24 |
-
The ideal value is 1.0, where acc=1 and mae=0.
|
25 |
-
"""
|
26 |
-
def __init__(self,name="Metric_Fun", **kwargs):
|
27 |
-
super(Metric_Fun,self).__init__(name=name, **kwargs)
|
28 |
-
self.evalue = self.add_weight('evalue', initializer='zeros')
|
29 |
-
# https://www.tensorflow.org/api_docs/python/tf/keras/metrics/BinaryAccuracy
|
30 |
-
self.acc = tf.keras.metrics.BinaryAccuracy()
|
31 |
-
# https://www.tensorflow.org/api_docs/python/tf/keras/metrics/MeanAbsoluteError
|
32 |
-
self.mae = tf.keras.metrics.MeanAbsoluteError()
|
33 |
-
|
34 |
-
def update_state(self, y_true, y_pred, sample_weight=None):
|
35 |
-
|
36 |
-
y_true = tf.cast(y_true,dtype=tf.float32)
|
37 |
-
y_pred = tf.cast(y_pred, dtype=tf.float32)
|
38 |
-
|
39 |
-
self.mae.update_state(y_true[:,4:5], y_pred[:,4:5])
|
40 |
-
abs_error = self.mae.result()
|
41 |
-
|
42 |
-
self.acc.update_state(y_true[ : , 0:4], y_pred[ : , 0:4])
|
43 |
-
accracy = self.acc.result()
|
44 |
-
|
45 |
-
#evalue = accracy
|
46 |
-
evalue = accracy - abs_error
|
47 |
-
self.evalue.assign(evalue)
|
48 |
-
|
49 |
-
def result(self):
|
50 |
-
return self.evalue
|
51 |
-
|
52 |
-
def reset_state(self):
|
53 |
-
# The state of the metric will be reset at the start of each epoch.
|
54 |
-
self.evalue.assign(0.0)
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
class GasNet3:
|
60 |
-
"""
|
61 |
-
Initialize, setting the input pixel, strat wavelength, end wavelength, and output channel
|
62 |
-
and the network name
|
63 |
-
"""
|
64 |
-
def __init__(self,Network_name,Output_channel):
|
65 |
-
#self.Input_pixel = 10000
|
66 |
-
#self.Start_wavelength = 4000
|
67 |
-
#self.End_wavelength = 9000
|
68 |
-
#self.Input_wavelength = np.linspace(self.Start_wavelength,self.End_wavelength,self.Input_pixel)
|
69 |
-
|
70 |
-
self.Network_name = Network_name
|
71 |
-
self.Input_wavelength = np.load('./test_data/wavelengths.npy')
|
72 |
-
self.Input_pixel = len(self.Input_wavelength)
|
73 |
-
self.Inpt = tf.keras.layers.Input(shape=(self.Input_pixel,1)) #shape of spectra
|
74 |
-
self.Output_channel = Output_channel
|
75 |
-
self.batch = 128 # training batch
|
76 |
-
self.redshift_range = [0,4]
|
77 |
-
self.class_names = {b'AGN':0,b'GALAXY':1,b'QSO':2,b'STAR':3}
|
78 |
-
self.lable_dim = len(self.class_names)
|
79 |
-
|
80 |
-
def Wavelength_Grid(self):
|
81 |
-
"""
|
82 |
-
Return the grid of input wavelength
|
83 |
-
"""
|
84 |
-
return self.Input_wavelength
|
85 |
-
|
86 |
-
def Interpolate_Flux(self,wavelength,flux):
|
87 |
-
"""
|
88 |
-
Interpolate the specturm flux into a suitable shape
|
89 |
-
"""
|
90 |
-
if flux.ndim != 1:
|
91 |
-
Int_flux = [np.interp(self.Input_wavelength,wavelength[i],flux[i]) for i in range(len(flux))]
|
92 |
-
Int_flux = np.array(Int_flux)
|
93 |
-
else:
|
94 |
-
Int_flux = np.interp(self.Input_wavelength,wavelength,flux)
|
95 |
-
return Int_flux
|
96 |
-
|
97 |
-
|
98 |
-
def Append_Noise_Sample(self):
|
99 |
-
"""
|
100 |
-
a extra blank noise will add during training
|
101 |
-
"""
|
102 |
-
pass
|
103 |
-
|
104 |
-
|
105 |
-
def Block_ResNet(self,x0,n):
|
106 |
-
"""
|
107 |
-
one ResNet Block, to reduce feature dimension
|
108 |
-
"""
|
109 |
-
core_size = 5
|
110 |
-
x=tf.keras.layers.Conv1D(n,kernel_size=core_size,strides=2,padding='same')(x0)
|
111 |
-
x=tf.keras.layers.BatchNormalization()(x)
|
112 |
-
x=tf.keras.layers.Activation('relu')(x)
|
113 |
-
x=tf.keras.layers.Conv1D(2*n, kernel_size=core_size,padding='same')(x)
|
114 |
-
x=tf.keras.layers.BatchNormalization()(x)
|
115 |
-
ShortCut = tf.keras.layers.Conv1D(2*n,kernel_size=2,strides=2,padding='same')(x0)
|
116 |
-
x = tf.keras.layers.Add()([x,ShortCut])
|
117 |
-
x=tf.keras.layers.Activation('relu')(x)
|
118 |
-
x = tf.keras.layers.MaxPooling1D(pool_size=core_size,strides=2)(x)
|
119 |
-
return x
|
120 |
-
|
121 |
-
def Block_ResNet_2(self,x0,n):
|
122 |
-
"""
|
123 |
-
one ResNet Block, to not reduce feature dimension, but extend channels.
|
124 |
-
"""
|
125 |
-
core_size = 3
|
126 |
-
x=tf.keras.layers.Conv1D(n,kernel_size=core_size,strides=1,padding='same')(x0)
|
127 |
-
x=tf.keras.layers.BatchNormalization()(x)
|
128 |
-
x=tf.keras.layers.Activation('relu')(x)
|
129 |
-
x=tf.keras.layers.Conv1D(n,kernel_size=core_size,strides=1,padding='same')(x)
|
130 |
-
x=tf.keras.layers.BatchNormalization()(x)
|
131 |
-
x=tf.keras.layers.Activation('relu')(x)
|
132 |
-
x=tf.keras.layers.Conv1D(2*n, kernel_size=core_size,strides=1,padding='same')(x)
|
133 |
-
x=tf.keras.layers.BatchNormalization()(x)
|
134 |
-
ShortCut = tf.keras.layers.Conv1D(2*n,kernel_size=1,strides=1,padding='same')(x0)
|
135 |
-
x = tf.keras.layers.Add()([x,ShortCut])
|
136 |
-
x=tf.keras.layers.Activation('relu')(x)
|
137 |
-
return x
|
138 |
-
|
139 |
-
def ResNet(self,x):
|
140 |
-
"""
|
141 |
-
Networks made by Blocks
|
142 |
-
"""
|
143 |
-
x = self.Block_ResNet(x,16)
|
144 |
-
x = self.Block_ResNet(x,32)
|
145 |
-
x = self.Block_ResNet(x,64)
|
146 |
-
x = self.Block_ResNet(x,128)
|
147 |
-
x = self.Block_ResNet(x,256)
|
148 |
-
#x = self.Block_ResNet(x,512)
|
149 |
-
#x = self.Block_ResNet(x,1024)
|
150 |
-
x = tf.keras.layers.Flatten()(x)
|
151 |
-
x = tf.keras.layers.Dense(1024, activation='relu')(x)
|
152 |
-
x = tf.keras.layers.Dense(self.Output_channel,activation=None)(x)
|
153 |
-
x0 = tf.keras.layers.Activation('softmax')(x[ : , 0: self.lable_dim])
|
154 |
-
x1 = x[ : , self.lable_dim : self.Output_channel]
|
155 |
-
x = tf.keras.layers.Concatenate(axis=-1)([x0, x1])
|
156 |
-
return x
|
157 |
-
|
158 |
-
def ResNet_test(self,x):
|
159 |
-
"""
|
160 |
-
Networks for testing
|
161 |
-
"""
|
162 |
-
x = self.Block_ResNet(x,16)
|
163 |
-
x = self.Block_ResNet(x,32)
|
164 |
-
x = self.Block_ResNet(x,64)
|
165 |
-
x = self.Block_ResNet(x,128)
|
166 |
-
x = self.Block_ResNet_2(x,256)
|
167 |
-
x = self.Block_ResNet_2(x,512)
|
168 |
-
x = self.Block_ResNet_2(x,1024)
|
169 |
-
|
170 |
-
x = tf.keras.layers.Flatten()(x)
|
171 |
-
x = tf.keras.layers.Dense(1024, activation='relu')(x)
|
172 |
-
#x = tf.keras.layers.Dropout(0.4)(x)
|
173 |
-
x = tf.keras.layers.Dense(self.Output_channel,activation=None)(x)
|
174 |
-
x0 = tf.keras.layers.Activation('softmax')(x[ : , 0: self.lable_dim])
|
175 |
-
x1 = x[ : , self.lable_dim : self.Output_channel]
|
176 |
-
x = tf.keras.layers.Concatenate(axis=-1)([x0, x1])
|
177 |
-
return x
|
178 |
-
|
179 |
-
def Built_Model(self,test=False):
|
180 |
-
"""
|
181 |
-
Return the ResNet mdoels
|
182 |
-
"""
|
183 |
-
if test:
|
184 |
-
model = tf.keras.Model(inputs=self.Inpt,outputs=self.ResNet_test(self.Inpt),name=self.Network_name)
|
185 |
-
else:
|
186 |
-
model = tf.keras.Model(inputs=self.Inpt,outputs=self.ResNet(self.Inpt),name=self.Network_name)
|
187 |
-
model.summary()
|
188 |
-
return model
|
189 |
-
|
190 |
-
def Plot_Model(self,test=False):
|
191 |
-
"""
|
192 |
-
Plot the network architecture
|
193 |
-
"""
|
194 |
-
model = self.Built_Model(test)
|
195 |
-
tf.keras.utils.plot_model(model,to_file=model.name+'.pdf',show_shapes=True,show_layer_names=False)
|
196 |
-
|
197 |
-
def Data_Clip(self,label,redshift):
|
198 |
-
"""
|
199 |
-
Conevrt the label to one-hot code.
|
200 |
-
Redshfit are set on a range.
|
201 |
-
Contact them into a vector.
|
202 |
-
"""
|
203 |
-
# reshape the label and reshift array
|
204 |
-
label = np.array(label)
|
205 |
-
redshift = np.array(redshift)
|
206 |
-
redshift = redshift.reshape(len(redshift),1)
|
207 |
-
# convert to one-hot coded
|
208 |
-
value = np.vectorize(self.class_names.get)(label)
|
209 |
-
label = tf.keras.utils.to_categorical(value, num_classes=self.lable_dim)
|
210 |
-
redshift = np.clip(redshift, self.redshift_range[0]-1, self.redshift_range[1]+1)
|
211 |
-
redshift = tf.convert_to_tensor(redshift)
|
212 |
-
vector = tf.concat([label,redshift],axis=-1) # a veter made by label and redshift concation
|
213 |
-
return vector
|
214 |
-
|
215 |
-
def Preprocess(self,flux):
|
216 |
-
"""
|
217 |
-
The input flux and label should be propocess
|
218 |
-
"""
|
219 |
-
#flux = flux - np.mean(flux,-1)
|
220 |
-
flux = tf.keras.utils.normalize(flux,axis=-1) # flux/sum(flux**2)**0.5
|
221 |
-
# https://www.tensorflow.org/api_docs/python/tf/math/divide_no_nan
|
222 |
-
# flux = tf.math.divide_no_nan(flux, np.max(flux,axis=-1).reshape(flux.shape[0],1)) # Norm to 0-1
|
223 |
-
# flux = -np.log10(np.maximum(flux,0)+1e-26)
|
224 |
-
# flux = -np.log10(np.abs(flux)+1e-26)
|
225 |
-
# flux = np.clip(flux,0,4)
|
226 |
-
return flux
|
227 |
-
|
228 |
-
def Loss_Func(self,y_true,y_pred):
|
229 |
-
"""
|
230 |
-
The loss function of this models.
|
231 |
-
loss = absolute redshift error + label entroy
|
232 |
-
"""
|
233 |
-
# redshift_error
|
234 |
-
Huber = tf.keras.losses.Huber(0.01)
|
235 |
-
error = Huber(y_true[ : , self.lable_dim : self.Output_channel], y_pred[ : , self.lable_dim : self.Output_channel])
|
236 |
-
# entropy
|
237 |
-
Cce = tf.keras.losses.CategoricalCrossentropy()
|
238 |
-
crossentropy = Cce(y_true[ : , 0:self.lable_dim], y_pred[ : , 0:self.lable_dim])
|
239 |
-
#loss = crossentropy
|
240 |
-
loss = error + crossentropy
|
241 |
-
return loss
|
242 |
-
|
243 |
-
|
244 |
-
def Train_Model(self,data,lr=1e-3,epo=40,test=False):
|
245 |
-
"""
|
246 |
-
Training the model.
|
247 |
-
Input training data.
|
248 |
-
"""
|
249 |
-
batch = self.batch
|
250 |
-
if os.path.exists(self.Network_name+'.h5'):
|
251 |
-
model = tf.keras.models.load_model(self.Network_name+'.h5',custom_objects={'Loss_Func':self.Loss_Func,'Metric_Fun':Metric_Fun()})
|
252 |
-
print('loading the existed model')
|
253 |
-
else:
|
254 |
-
model = self.Built_Model(test)
|
255 |
-
optimizer = tf.keras.optimizers.Adam(learning_rate=lr) # Adam
|
256 |
-
model.compile(optimizer,loss=self.Loss_Func,metrics=Metric_Fun()) # complize model
|
257 |
-
# https://tensorflow.google.cn/api_docs/python/tf/keras/callbacks/ModelCheckpoint
|
258 |
-
checkPoint = tf.keras.callbacks.ModelCheckpoint(model.name+'.h5',monitor='val_Metric_Fun',mode='max',verbose=1,save_best_only=True,save_weights_only=False)# callback function
|
259 |
-
csvLogger = tf.keras.callbacks.CSVLogger(model.name+'.csv',append=True) # save training history
|
260 |
-
train_x, train_y = self.Preprocess(data['train']['flux']), self.Data_Clip(data['train']['label'],data['train']['redshift'])
|
261 |
-
valid_x, valid_y = self.Preprocess(data['valid']['flux']), self.Data_Clip(data['valid']['label'],data['valid']['redshift'])
|
262 |
-
model.fit(train_x,train_y,epochs=epo,batch_size=batch, validation_data=(valid_x,valid_y),callbacks=[checkPoint,csvLogger],shuffle=True)
|
263 |
-
|
264 |
-
def Prodiction(self,flux,lamb=[]):
|
265 |
-
"""
|
266 |
-
Predition, classes and redshift.
|
267 |
-
"""
|
268 |
-
if len(lamb) != 0:
|
269 |
-
flux = self.Interpolate_Flux(lamb,flux)
|
270 |
-
model = tf.keras.models.load_model(self.Network_name+'.h5',custom_objects={'Loss_Func':self.Loss_Func,'Metric_Fun':Metric_Fun()})
|
271 |
-
flux = self.Preprocess(flux)
|
272 |
-
pred = model.predict(flux) # give classes and reshift
|
273 |
-
|
274 |
-
pred_label,pred_redshift = np.hsplit(pred, [self.lable_dim])
|
275 |
-
pred_label = np.argmax(pred_label,axis=-1) # turn one-hot to integer value, get the max value index
|
276 |
-
dict = {v:k for k, v in self.class_names.items()} # reverse key and value of a dict
|
277 |
-
pred_label = np.vectorize(dict.get)(pred_label) # turn integer value to its name
|
278 |
-
|
279 |
-
return pred_label,pred_redshift
|
280 |
-
|
281 |
-
|
282 |
-
|
283 |
-
class Spec_Checker():
|
284 |
-
|
285 |
-
def __init__(self):
|
286 |
-
self.gasnet = GasNet3('test_net',Output_channel=5)
|
287 |
-
|
288 |
-
def Show_spec(self,lamb,flux,name=''):
|
289 |
-
"""
|
290 |
-
show the detail of spectra after interpolated and preprocessed
|
291 |
-
"""
|
292 |
-
plt.figure(figsize=(16,6),dpi=160)
|
293 |
-
int_flux = self.gasnet.Interpolate_Flux(lamb,flux)
|
294 |
-
|
295 |
-
plt.subplot(2,1,1)
|
296 |
-
plt.title(name + '-After interpolated')
|
297 |
-
plt.plot(lamb,flux,linewidth=0.5,label='original flux')
|
298 |
-
plt.plot(self.gasnet.Input_wavelength,int_flux,linewidth=0.5,label='interpolate flux')
|
299 |
-
plt.legend()
|
300 |
-
|
301 |
-
plt.subplot(2,1,2)
|
302 |
-
plt.title(name + '-After preprocessed')
|
303 |
-
plt.plot(lamb,self.gasnet.Preprocess(flux)[0],linewidth=0.5,label='original flux')
|
304 |
-
plt.plot(self.gasnet.Input_wavelength,self.gasnet.Preprocess(int_flux)[0],linewidth=0.5,label='interpolate flux')
|
305 |
-
plt.legend()
|
306 |
-
|
307 |
-
def SDSS_spec(self,file,plot=True):
|
308 |
-
"""
|
309 |
-
load the spectra from SDSS files
|
310 |
-
"""
|
311 |
-
data = Table.read(file)
|
312 |
-
flux, lamb = data['flux'], 10**data['loglam']
|
313 |
-
if plot:
|
314 |
-
self.Show_spec(lamb,flux, name='SDSS:' + file.rsplit('/')[-1])
|
315 |
-
spec_info = Table.read(file,2)
|
316 |
-
redshift, classes = spec_info['Z'][0], spec_info['CLASS'][0]
|
317 |
-
return {'wavelength':lamb,'flux':flux,'redshift':redshift,'label':classes}
|
318 |
-
|
319 |
-
def SDSS_spec_stack(self,num=0,plot=True):
|
320 |
-
"""
|
321 |
-
load the spectra of validation
|
322 |
-
"""
|
323 |
-
wavelength = self.gasnet.Input_wavelength
|
324 |
-
data = Table.read('train_data/val.fits')
|
325 |
-
flux,label,redshift = data['int_flux'],data['train_label'],data['Z']
|
326 |
-
wavelength = np.repeat([wavelength], len(flux), axis=0)
|
327 |
-
if plot:
|
328 |
-
self.Show_spec(wavelength[num],flux[num], name='validation:' + str(num))
|
329 |
-
return {'wavelength':wavelength,'flux':flux,'redshift':redshift,'label':label}
|
330 |
-
|
331 |
-
def JK_spec(self,file):
|
332 |
-
"""
|
333 |
-
load the spectrum files from JK mock
|
334 |
-
"""
|
335 |
-
data = Table.read(file)
|
336 |
-
flux, lamb = data['FLUX'][0], data['WAVE'][0]
|
337 |
-
self.Show_spec(lamb,flux, name='JK:' + file.rsplit('/')[-1])
|
338 |
-
|
339 |
-
def npy_file(self,num=0,plot=True):
|
340 |
-
"""
|
341 |
-
load the spectrum files from qcp test data
|
342 |
-
"""
|
343 |
-
wavelength = np.load('./test_data/wavelengths.npy')
|
344 |
-
flux = np.load('./test_data/data.npy')
|
345 |
-
wavelength = np.repeat([wavelength], len(flux), axis=0)
|
346 |
-
if plot:
|
347 |
-
self.Show_spec(wavelength[num],flux[num], name='test npy:' + str(num))
|
348 |
-
label = np.load('./test_data/labels.npy')
|
349 |
-
dict = {v:k for k, v in self.gasnet.class_names.items()} # reverse key and value of a dict
|
350 |
-
label = np.vectorize(dict.get)(label) # turn integer value to its name
|
351 |
-
return {'wavelength':wavelength,'flux':flux,'redshift':None,'label':label}
|
352 |
-
|
353 |
-
def Luke_spec(self,num=0,plot=True):
|
354 |
-
"""
|
355 |
-
load the spectrum files from Luck mock
|
356 |
-
"""
|
357 |
-
spec_file = '../Luke_mock_spectra/Luke_spec.fits' # flux need multiply 1e17
|
358 |
-
data = Table.read(spec_file)
|
359 |
-
wavelength = np.load('./test_data/wavelengths.npy')
|
360 |
-
wavelength = np.repeat([wavelength], len(data), axis=0)
|
361 |
-
flux = data['int_flux']
|
362 |
-
if plot:
|
363 |
-
self.Show_spec(wavelength[num],flux[num], name='Luck :' + str(num))
|
364 |
-
return {'wavelength':wavelength,'flux':flux,'redshift':data['Redshift'],'label':data['train_label']}
|
365 |
-
|
366 |
-
def JK_stack_spec(self,num=0,plot=True):
|
367 |
-
"""
|
368 |
-
load the spectrum files from Luck mock
|
369 |
-
"""
|
370 |
-
spec_file = './JK_stack_mock.fits'
|
371 |
-
data = Table.read(spec_file)
|
372 |
-
wavelength = Table.read('JK_mock_sample.fits')['WAVE'][0]
|
373 |
-
wavelength = np.repeat([wavelength], len(data), axis=0)
|
374 |
-
flux,label,redshift = data['FLUX'],data['train_type'],data['REDSHIFT']
|
375 |
-
if plot:
|
376 |
-
self.Show_spec(wavelength[num],flux[num], name='JK--num--'+str(num)+'--label--'+str(label[num])+'--redshift--'+str(redshift[num]))
|
377 |
-
return {'wavelength':wavelength,'flux':flux,'redshift':redshift,'label':label}
|
378 |
-
|
379 |
-
def Svae_Figure(self,data,name='test'):
|
380 |
-
"""
|
381 |
-
plot a serial of spectra in one pdf file
|
382 |
-
"""
|
383 |
-
figfile = 'figure'
|
384 |
-
if not os.path.exists(figfile):
|
385 |
-
os.mkdir(figfile)
|
386 |
-
fig, axes = plt.subplots(nrows=len(data['flux']),ncols=1,sharex=True,figsize=(8,2*len(data)),dpi=50)
|
387 |
-
fig.suptitle(name)
|
388 |
-
plt.xlabel('wavelength')
|
389 |
-
plt.ylabel('flux')
|
390 |
-
for i in range(len(data['flux'])):
|
391 |
-
axe = axes[i]
|
392 |
-
axe.plot(data['wavelength'][i],data['flux'][i],linewidth=0.5,label=data['label'][i]+' z='+str(data['redshift'][i]))
|
393 |
-
axe.legend()
|
394 |
-
fname = os.path.join(figfile,str(name)+'.pdf')
|
395 |
-
plt.savefig(fname)
|
396 |
-
plt.close()
|
397 |
-
|
398 |
-
def Confusion_Matrix(self,pred,real):
|
399 |
-
"""
|
400 |
-
plot the confusion matrix
|
401 |
-
"""
|
402 |
-
data = {'Actual':np.array(real).flatten(),'Predicted':np.array(pred).flatten()}
|
403 |
-
df = pd.DataFrame(data)
|
404 |
-
plt.figure(figsize=(8,6),dpi=160)
|
405 |
-
confusion_matrix = pd.crosstab(df['Actual'], df['Predicted'], rownames=['Actual'], colnames=['Predicted'])
|
406 |
-
sns.heatmap(confusion_matrix,cmap="crest", annot=True)
|
407 |
-
|
408 |
-
def One2One(self,pred,real,label):
|
409 |
-
"""
|
410 |
-
plot the redicted redshift vs. real
|
411 |
-
"""
|
412 |
-
data = {'pred_redshift':np.array(pred).flatten(),
|
413 |
-
'real_redshift':np.array(real).flatten(),
|
414 |
-
'label':np.array(label).flatten()}
|
415 |
-
df = pd.DataFrame(data)
|
416 |
-
# print(df.dtypes)
|
417 |
-
df['real_redshift'] = df['real_redshift'].astype('float32')
|
418 |
-
# https://seaborn.pydata.org/generated/seaborn.lmplot.html#seaborn.lmplot
|
419 |
-
sns.lmplot(data=df, x='pred_redshift', y='real_redshift', hue='label',col='label',
|
420 |
-
col_wrap=2, height=6, #plot size
|
421 |
-
line_kws={"alpha":0.1}, #ci=None, #line style
|
422 |
-
scatter_kws={"s":1,"alpha":1},sharex=False, sharey=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|