def output_error_graph(output_dim_,Y_header,Date_array,Y,Y_pred,name,
fsize = [10,8],color = ['r','g','b'],size = 1,\
filetype_s = '.jpg'):
fig = plt.figure(figsize=fsize)
error = (Y_pred-Y)**2
for i in range(0,output_dim_,1):
plt.subplot(output_dim_,1 , i+1)
plt.title(Y_header[i])
plt.scatter(Date_array,error[:,i],c = color[i],s = size)
plt.tight_layout()
export_graphs(name,fig,filetype = filetype_s)
def output_graphs(output_dim_,Y_header,Y_valid,Y_pred,name,filetype_s = '.jpg',\
x_fig = 8,y_fig = 3,color = ['r','g','b'],size = 0.05,\
parity=True,limits=True,\
xlimit = np.array([-0.1,1.1]),ylimit = np.array([-0.1,1.1])):
xy_parity = np.linspace(0,1,100)
fig = plt.figure(figsize=(x_fig,y_fig))
for i in range(0,output_dim_,1):
plt.subplot(1, output_dim_, i+1)
plt.title(Y_header[i])
plt.scatter(Y_valid[:,i],Y_pred[:,i],label = Y_header[i],c = color[i],s = size)
if parity:
plt.plot(xy_parity,xy_parity,'k')
if limits:
plt.xlim(xlimit[0],xlimit[1])
plt.ylim(ylimit[0],ylimit[1])
plt.tight_layout()
export_graphs(name,fig,filetype = filetype_s)
max_error_ppb = np.max(error_ppb)
min_error_ppb = np.min(error_ppb)
#
for i in range(0, len(Sensor_name), 1):
sensor_index = indexall(Sensor_List, Sensor_name[i])
#plot datetime vs abs error
fig = plt.figure()
plt.plot(date[sensor_index],\
error[sensor_index],'-k',linewidth=2.0)
plt.xlabel('Date')
plt.ylabel('Error')
plt.title('Time Series of Sensor:' + Sensor_name[i])
plt.tight_layout(pad=0.1)
name = 'Time_Series' + Sensor_name[i]
export_graphs(name, fig, filetype='.jpg')
#plot datetime vs error in ppm
fig = plt.figure()
plt.plot(date[sensor_index],\
error_ppb[sensor_index],'-k',linewidth=2.0)
plt.xlabel('Date')
plt.ylabel('Error ($ppb$)')
plt.title('Real Variable Time Series of Sensor:' + Sensor_name[i])
plt.tight_layout(pad=0.1)
name = 'Real_Variable_Time_Series' + Sensor_name[i]
export_graphs(name, fig, filetype='.jpg')
#plot datetime vs Ozone in ppm
Stack_Predict[:, Y_header_list] = Y_pred
#Inverse both the predicted and the actual data
Data_Array_Actual_invnorm = s.inverse_transform(Stack_Actual)
Data_Array_Predict_invnorm = s.inverse_transform(Stack_Predict)
#Store Y data
Y_pred_ppm = Data_Array_Predict_invnorm[:, Y_Loc] * 1000
Y_act_ppm = Data_Array_Actual_invnorm[:, Y_Loc] * 1000
#plot the parity plot for the normalized validation data
fig = plot_parity(Y_valid[:, 0],
Y_pred[:, 0],
'Normalized $O_3$ Values',
'Predicted $O_3$ Values',
s=0.5)
export_graphs('norm_parity_valid', fig, filetype='.jpg')
#plot the parity plot for the unnormalized validation data
fig = plot_parity(Y_act_ppm,
Y_pred_ppm,
'Actual $O_3$ Values($ppb$)',
'Predicted $O_3$ Values($ppb$)',
s=0.5,
maxset=True)
export_graphs('unnorm_parity_valid', fig, filetype='.jpg')
#load in the outside dataset
df_test = pd.read_csv(outside_data_path)
data_array_test = np.array(df_test)
X_Other = data_array_test[:, X_header_list]
Y_Other = data_array_test[:, Y_header_list]
#read data and parse the data
df = pd.read_csv(full_path)
header =list(df)
header_num = len(header)
#get X and Y values
Full_List = list(np.arange(0,header_num-1+1e-6,1,dtype=int))
Y_Loc = header.index('O3 Value')
Y_header_list = []
Y_header_list.append(Y_Loc)
X_header_list = list(set(Full_List)-set(Y_header_list))
#load neural network
load_model_path =\
r"E:\PhD project\ozone\Saved_Neural_Networks\082119_nnet_node_mse_40layer_1new.h5"
model = load_model(load_model_path,custom_objects={'r2_keras':r2_keras})
model.summary()
#call custom function
fig_save,plt_name_save,range_save = Parameter_Analysis(model,header,X_header_list)
#export figure
for i in range(0,len(fig_save)):
export_graphs(plt_name_save[i],fig_save[i])
#plot hist
#X_header = header.remove('O3 Value')
#plt.figure()
#plt.bar(range_save,label=X_header)