def create_image_training_epoch(X_data_tr, Y_data_tr, X_data_val, Y_data_val, tr_loss, val_loss, x_grid, y_grid, cf_a, video_fotograms_folder, epoch_i): """ Creates the image of the training and validation accuracy """ gl.init_figure(); ax1 = gl.subplot2grid((2,1), (0,0), rowspan=1, colspan=1) ax2 = gl.subplot2grid((2,1), (1,0), rowspan=1, colspan=1) plt.title("Training") ## First plot with the data and predictions !!! ax1 = gl.scatter(X_data_tr, Y_data_tr, ax = ax1, lw = 3,legend = ["tr points"], labels = ["Analysis of training", "X","Y"]) gl.scatter(X_data_val, Y_data_val, lw = 3,legend = ["val points"]) gl.plot (x_grid, y_grid, legend = ["Prediction function"]) gl.set_zoom(xlimPad = [0.2, 0.2], ylimPad = [0.2,0.2], X = X_data_tr, Y = Y_data_tr) ## Second plot with the evolution of parameters !!! ax2 = gl.plot([], tr_loss, ax = ax2, lw = 3, labels = ["RMSE. lr: %.3f"%cf_a.lr, "epoch","RMSE"], legend = ["train"]) gl.plot([], val_loss, lw = 3, legend = ["validation"], loc = 3) gl.set_fontSizes(ax = [ax1,ax2], title = 20, xlabel = 20, ylabel = 20, legend = 20, xticks = 12, yticks = 12) # Set final properties and save figure gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.30) gl.savefig(video_fotograms_folder +'%i.png'%epoch_i, dpi = 100, sizeInches = [14, 10], close = True, bbox_inches = None)
def plot_weights_network(model, folder_images): # weights = model.linear1.weight.detach().numpy() biases = model.linear1.bias.detach().numpy().reshape(-1,1) neurons = np.concatenate((weights, biases), axis = 1) weights2 = model.W2.detach().numpy() biases2 = model.b2.detach().numpy().reshape(-1,1) neurons2 = np.concatenate((weights2, biases2), axis =0).T gl.init_figure(); ax1 = gl.subplot2grid((1,4), (0,0), rowspan=1, colspan=2) ax2 = gl.subplot2grid((1,4), (0,3), rowspan=1, colspan=4) cmap = cm.get_cmap('coolwarm', 30) cax = ax1.imshow(neurons, interpolation="nearest", cmap=cmap) cax2 = ax2.imshow(neurons2, interpolation="nearest", cmap=cmap) # plt.xticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='vertical') # plt.yticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='horizontal') plt.colorbar(cax) # plt.colorbar(cax2) # ax1.set_xticks(data_df_train.columns) # , rotation='vertical' # ax1.grid(True) plt.title('Weights ') # labels=[str(x) for x in range(Nshow )] # ax1.set_xticklabels(labels,fontsize=20) # ax1.set_yticklabels(labels,fontsize=20) # Add colorbar, make sure to specify tick locations to match desired ticklabels plt.show() gl.savefig(folder_images +'Weights.png', dpi = 100, sizeInches = [2*8, 2*2])
def plot_weights_network(model, folder_images): # weights = model.linear1.weight.detach().numpy() biases = model.linear1.bias.detach().numpy().reshape(-1, 1) neurons = np.concatenate((weights, biases), axis=1) weights2 = model.W2.detach().numpy() biases2 = model.b2.detach().numpy().reshape(-1, 1) neurons2 = np.concatenate((weights2, biases2), axis=0).T gl.init_figure() ax1 = gl.subplot2grid((1, 4), (0, 0), rowspan=1, colspan=2) ax2 = gl.subplot2grid((1, 4), (0, 3), rowspan=1, colspan=4) cmap = cm.get_cmap('coolwarm', 30) cax = ax1.imshow(neurons, interpolation="nearest", cmap=cmap) cax2 = ax2.imshow(neurons2, interpolation="nearest", cmap=cmap) # plt.xticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='vertical') # plt.yticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='horizontal') plt.colorbar(cax) # plt.colorbar(cax2) # ax1.set_xticks(data_df_train.columns) # , rotation='vertical' # ax1.grid(True) plt.title('Weights ') # labels=[str(x) for x in range(Nshow )] # ax1.set_xticklabels(labels,fontsize=20) # ax1.set_yticklabels(labels,fontsize=20) # Add colorbar, make sure to specify tick locations to match desired ticklabels plt.show() gl.savefig(folder_images + 'Weights.png', dpi=100, sizeInches=[2 * 8, 2 * 2])
def plot_learnt_function(X_data_tr, Y_data_tr, X_data_val, Y_data_val, x_grid, y_grid, cf_a, folder_images): gl.init_figure() ax1 = gl.scatter(X_data_tr, Y_data_tr, lw=3, legend=["tr points"], labels=["Data", "X", "Y"], alpha=0.2) ax2 = gl.scatter(X_data_val, Y_data_val, lw=3, legend=["val points"], alpha=0.2) gl.set_fontSizes(ax=[ax1, ax2], title=20, xlabel=20, ylabel=20, legend=20, xticks=12, yticks=12) gl.plot(x_grid, y_grid, legend=["training line"]) gl.savefig(folder_images + 'Training_Example_Data.png', dpi=100, sizeInches=[14, 4])
def create_image_training_epoch(X_data_tr, Y_data_tr, X_data_val, Y_data_val, tr_loss, val_loss, x_grid, y_grid, cf_a, video_fotograms_folder, epoch_i): """ Creates the image of the training and validation accuracy """ gl.init_figure() ax1 = gl.subplot2grid((2, 1), (0, 0), rowspan=1, colspan=1) ax2 = gl.subplot2grid((2, 1), (1, 0), rowspan=1, colspan=1) plt.title("Training") ## First plot with the data and predictions !!! ax1 = gl.scatter(X_data_tr, Y_data_tr, ax=ax1, lw=3, legend=["tr points"], labels=["Analysis of training", "X", "Y"]) gl.scatter(X_data_val, Y_data_val, lw=3, legend=["val points"]) gl.plot(x_grid, y_grid, legend=["Prediction function"]) gl.set_zoom(xlimPad=[0.2, 0.2], ylimPad=[0.2, 0.2], X=X_data_tr, Y=Y_data_tr) ## Second plot with the evolution of parameters !!! ax2 = gl.plot([], tr_loss, ax=ax2, lw=3, labels=["RMSE. lr: %.3f" % cf_a.lr, "epoch", "RMSE"], legend=["train"]) gl.plot([], val_loss, lw=3, legend=["validation"], loc=3) gl.set_fontSizes(ax=[ax1, ax2], title=20, xlabel=20, ylabel=20, legend=20, xticks=12, yticks=12) # Set final properties and save figure gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.30) gl.savefig(video_fotograms_folder + '%i.png' % epoch_i, dpi=100, sizeInches=[14, 10], close=True, bbox_inches=None)
def plot_evolution_RMSE(tr_loss, val_loss, cf_a, folder_images): gl.init_figure() ax1 = gl.plot([], tr_loss, lw = 3, labels = ["RMSE loss and parameters. Learning rate: %.3f"%cf_a.lr, "","RMSE"], legend = ["train"]) gl.plot([], val_loss, lw = 3, legend = ["validation"]) gl.set_fontSizes(ax = [ax1], title = 20, xlabel = 20, ylabel = 20, legend = 20, xticks = 12, yticks = 12) gl.savefig(folder_images +'Training_Example_Parameters.png', dpi = 100, sizeInches = [14, 7])
def plot_learnt_function(X_data_tr, Y_data_tr, X_data_val, Y_data_val, x_grid, y_grid, cf_a, folder_images): gl.init_figure() ax1 = gl.scatter(X_data_tr, Y_data_tr, lw = 3,legend = ["tr points"], labels = ["Data", "X","Y"], alpha = 0.2) ax2 = gl.scatter(X_data_val, Y_data_val, lw = 3,legend = ["val points"], alpha = 0.2) gl.set_fontSizes(ax = [ax1,ax2], title = 20, xlabel = 20, ylabel = 20, legend = 20, xticks = 12, yticks = 12) gl.plot (x_grid, y_grid, legend = ["training line"]) gl.savefig(folder_images +'Training_Example_Data.png', dpi = 100, sizeInches = [14, 4])
def plot_evolution_RMSE(tr_loss, val_loss, cf_a, folder_images): gl.init_figure() ax1 = gl.plot([], tr_loss, lw=3, labels=[ "RMSE loss and parameters. Learning rate: %.3f" % cf_a.lr, "", "RMSE" ], legend=["train"]) gl.plot([], val_loss, lw=3, legend=["validation"]) gl.set_fontSizes(ax=[ax1], title=20, xlabel=20, ylabel=20, legend=20, xticks=12, yticks=12) gl.savefig(folder_images + 'Training_Example_Parameters.png', dpi=100, sizeInches=[14, 7])
def create_image_weights_epoch(model, video_fotograms_folder2, epoch_i): """ Creates the image of the training and validation accuracy """ N_Bayesian_layers = len(model.VBmodels) N_Normal_layers = len(model.LinearModels) # Compute the number of squares we will need: # 1 x linear layers, 2 x LSTMS gl.init_figure() cmap = cm.get_cmap('coolwarm', 30) all_axes = [] for i in range(N_Bayesian_layers): layer = model.VBmodels[i] # if (layer.type_layer == "linear"): if ("linear" in type(layer).__name__.lower()): ax = gl.subplot2grid((1, N_Bayesian_layers + N_Normal_layers), (0, i), rowspan=1, colspan=1) weights = layer.weight.detach().cpu().numpy() biases = layer.bias.detach().cpu().numpy().reshape(-1, 1) neurons = np.concatenate((weights, biases), axis=1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) else: ax = gl.subplot2grid((1, N_Bayesian_layers + N_Normal_layers), (0, i), rowspan=1, colspan=1) weights_ih = layer.weight_ih.detach().cpu().numpy() biases_ih = layer.bias_ih.detach().cpu().numpy().reshape(-1, 1) weights_hh = layer.weight_hh.detach().cpu().numpy() biases_hh = layer.bias_hh.detach().cpu().numpy().reshape(-1, 1) weights = np.concatenate((weights_ih, weights_hh), axis=1) biases = np.concatenate((biases_ih, biases_hh), axis=1) neurons = np.concatenate((weights, biases), axis=1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) for i in range(N_Normal_layers): layer = model.LinearModels[i] if ("linear" in type(layer).__name__.lower()): ax = gl.subplot2grid((1, N_Bayesian_layers + N_Normal_layers), (0, N_Bayesian_layers + i), rowspan=1, colspan=1) weights = layer.weight.detach().cpu().numpy() biases = layer.bias.detach().cpu().numpy().reshape(-1, 1) neurons = np.concatenate((weights, biases), axis=1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) else: ax = gl.subplot2grid((1, N_Bayesian_layers + N_Normal_layers), (0, N_Bayesian_layers + i), rowspan=1, colspan=1) weights_ih = layer.weight_ih.detach().cpu().numpy() biases_ih = layer.bias_ih.detach().cpu().numpy().reshape(-1, 1) weights_hh = layer.weight_hh.detach().cpu().numpy() biases_hh = layer.bias_hh.detach().cpu().numpy().reshape(-1, 1) weights = np.concatenate((weights_ih, weights_hh), axis=1) biases = np.concatenate((biases_ih, biases_hh), axis=1) neurons = np.concatenate((weights, biases), axis=1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) # plt.xticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='vertical') # plt.yticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='horizontal') plt.colorbar(cax) # plt.colorbar(cax2) # ax1.set_xticks(data_df_train.columns) # , rotation='vertical' # ax1.grid(True) plt.title('Weights ') # labels=[str(x) for x in range(Nshow )] # ax1.set_xticklabels(labels,fontsize=20) # ax1.set_yticklabels(labels,fontsize=20) # Add colorbar, make sure to specify tick locations to match desired ticklabels plt.show() gl.set_fontSizes(ax=[all_axes], title=20, xlabel=20, ylabel=20, legend=20, xticks=12, yticks=12) # Set final properties and save figure gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.30) gl.savefig(video_fotograms_folder2 + '%i.png' % epoch_i, dpi=100, sizeInches=[14, 10], close=True, bbox_inches=None)
def create_Bayesian_analysis_charts(model, X_data_tr, X_data_val, tr_data_loss, val_data_loss, KL_loss_tr, KL_loss_val, final_loss_tr, final_loss_val, folder_images, epoch_i=None): # Configurations of the plots alpha_points = 0.2 color_points_train = "dark navy blue" color_points_val = "amber" color_truth = "k" color_mean = "b" color_most_likey = "y" ################################ Divide in plots ############################## gl.init_figure() ax1 = gl.subplot2grid((6, 3), (0, 0), rowspan=3, colspan=1) ax2 = gl.subplot2grid((6, 3), (3, 0), rowspan=3, colspan=1, sharex=ax1) ax3 = gl.subplot2grid((6, 3), (0, 1), rowspan=2, colspan=1) ax4 = gl.subplot2grid((6, 3), (2, 1), rowspan=2, colspan=1, sharex=ax3) ax5 = gl.subplot2grid((6, 3), (4, 1), rowspan=2, colspan=1, sharex=ax3) ax6 = gl.subplot2grid((6, 3), (0, 2), rowspan=3, colspan=1) ax7 = gl.subplot2grid((6, 3), (3, 2), rowspan=3, colspan=1, sharex=ax6) """ ############################# Data computation ####################### """ Xtrain_sample_cpu, Xtrain_reconstruction,Xtrain_reconstruction_samples = \ compute_reconstruction_data( model,X_data_tr, Nsamples = 100, sample_index = 2) plot_reconstruction_data(Xtrain_sample_cpu, Xtrain_reconstruction, Xtrain_reconstruction_samples, ax1, ax2) """ ############## ax3 ax4 ax5: Loss Evolution !! ###################### """ plot_losses_evolution_epoch(tr_data_loss, val_data_loss, KL_loss_tr, KL_loss_val, final_loss_tr, final_loss_val, ax3, ax4, ax5) """ ############## ax6 ax7: Projecitons Weights !! ###################### """ plot_projections_VAE(model, X_data_tr, ax6) ## Plot in chart 7 the acceptable mu = 2sigma -> sigma = |mu|/2sigma # gl.set_zoom (ax = ax6, ylim = [-0.1,10]) # gl.set_zoom (ax = ax7, xlim = [-2.5, 2.5], ylim = [-0.05, np.exp(model.cf_a.input_layer_prior["log_sigma2"])*(1 + 0.15)]) # gl.set_zoom (ax = ax7, xlim = [-2.5, 2.5], ylim = [-0.1,2]) # Set final properties and save figure gl.set_fontSizes(ax=[ax1, ax2, ax3, ax4, ax5, ax6, ax7], title=20, xlabel=20, ylabel=20, legend=10, xticks=12, yticks=12) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.10) if (type(epoch_i) == type(None)): gl.savefig(folder_images + "../" + 'Final_values_regression_1D_' + str(model.cf_a.eta_KL) + '.png', dpi=100, sizeInches=[20, 10]) else: gl.savefig(folder_images + '%i.png' % epoch_i, dpi=100, sizeInches=[20, 10], close=True, bbox_inches="tight")
## Plot the covariance matrix ! # Show the Nshow first samples Nshow = 20 ## Plot realizations of the Gaussian process Nrealizations = 10 flag = 1; legend = ["Realizations"] labels = ["Gaussian Process noise e(t)","t", "e(t)"] # Plot the realizations gl.init_figure(); ax0 = gl.subplot2grid((1,4), (0,0), rowspan=1, colspan=3) for i in range(Nrealizations): f_prime = np.random.randn(N,1) error = L.dot(f_prime) gl.plot(tgrid,error, lw = 3, color = "b", ls = "-", alpha = 0.5, legend = legend, labels = labels) # gl.scatter(tgrid,f_prime, lw = 1, alpha = 0.3, color = "b") if (flag == 1): flag = 0 legend = [] #Variance of each prediction v = np.diagonal(K)
def plot_multiple_iterations(Xs,mus,covs, Ks ,myDManager, logl,theta_list,model_theta_list, folder_images): ######## Plot the original data ##### gl.init_figure(); gl.set_subplots(2,3); Ngraph = 6 colors = ["r","b","g"] K_G,K_W,K_vMF = Ks for i in range(Ngraph): indx = int(i*((len(theta_list)-1)/float(Ngraph-1))) nf = 1 for xi in range(len( Xs)): ## First cluster labels = ['EM Evolution. Kg:'+str(K_G)+ ', Kw:' + str(K_W) + ', K_vMF:' + str(K_vMF), "X1","X2"] ax1 = gl.scatter(Xs[xi][0,:],Xs[xi][1,:],labels = ["","",""] , color = colors[xi] ,alpha = 0.2, nf = nf) nf =0 mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = mus[xi], Sigma = covs[xi], Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "--", lw = 2 ,AxesStyle = "Normal2", color = colors[xi], alpha = 0.7) # Only doable if the clusters dont die for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): ## Plot the ecolution of the mu #### Plot the Covariance of the clusters ! mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = theta_list[indx][k][0], Sigma = theta_list[indx][k][1], Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "-.", lw = 3, AxesStyle = "Normal2", legend = ["Kg(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) elif(distribution_name == "Watson"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2*np.pi, Nsa) Xgrid= np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append(np.exp(Wad.Watson_pdf_log(Xgrid[:,i],[mu,kappa]) )) probs = np.array(probs) # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) gl.plot(X1_w,X2_w, alpha = 1, lw = 3, ls = "-.",legend = ["Kw(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) elif(distribution_name == "vonMisesFisher"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]); mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2*np.pi, Nsa) Xgrid= np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append(np.exp(vMFd.vonMisesFisher_pdf_log(Xgrid[:,i],[mu,kappa]) )) probs = np.array(probs) probs = probs.reshape((probs.size,1)).T # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) # print X1_w.shape, X2_w.shape gl.plot(X1_w,X2_w, alpha = 1, lw = 3, ls = "-.", legend = ["Kvmf(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) ax1.axis('equal') gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.2, hspace=0.01) gl.savefig(folder_images +'Final_State2. K_G:'+str(K_G)+ ', K_W:' + str(K_W) + '.png', dpi = 100, sizeInches = [18, 8])
def generate_images_iterations_ll(Xs,mus,covs, Ks ,myDManager, logl,theta_list,model_theta_list,folder_images_gif): # os.remove(folder_images_gif) # Remove previous images if existing """ WARNING: MEANT FOR ONLY 3 Distributions due to the color RGB """ import shutil ul.create_folder_if_needed(folder_images_gif) shutil.rmtree(folder_images_gif) ul.create_folder_if_needed(folder_images_gif) ######## Plot the original data ##### Xdata = np.concatenate(Xs,axis = 1).T colors = ["r","b","g"] K_G,K_W,K_vMF = Ks ### FOR EACH ITERATION for i in range(len(theta_list)): # theta_list indx = i gl.init_figure() ax1 = gl.subplot2grid((1,2), (0,0), rowspan=1, colspan=1) ## Get the relative ll of the Gaussian denoising cluster. ll = myDManager.pdf_log_K(Xdata,theta_list[indx]) N,K = ll.shape # print ll.shape for j in range(N): # For every sample #TODO: Can this not be done without a for ? # Normalize the probability of the sample being generated by the clusters Marginal_xi_probability = gf.sum_logs(ll[j,:]) ll[j,:] = ll[j,:]- Marginal_xi_probability ax1 = gl.scatter(Xdata[j,0],Xdata[j,1], labels = ['EM Evolution. Kg:'+str(K_G)+ ', Kw:' + str(K_W) + ', K_vMF:' + str(K_vMF), "X1","X2"], color = (np.exp(ll[j,1]), np.exp(ll[j,0]), np.exp(ll[j,2])) , ### np.exp(ll[j,2]) alpha = 1, nf = 0) # Only doable if the clusters dont die for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): ## Plot the ecolution of the mu #### Plot the Covariance of the clusters ! mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = theta_list[indx][k][0], Sigma = theta_list[indx][k][1], Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "-.", lw = 3, AxesStyle = "Normal2", legend = ["Kg(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) elif(distribution_name == "Watson"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]); mu = theta_list[-1][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2*np.pi, Nsa) Xgrid= np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append(np.exp(Wad.Watson_pdf_log(Xgrid[:,i],[mu,kappa]) )) probs = np.array(probs) # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) gl.plot(X1_w,X2_w, alpha = 1, lw = 3, ls = "-.", legend = ["Kw(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) elif(distribution_name == "vonMisesFisher"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]); mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2*np.pi, Nsa) Xgrid= np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append(np.exp(vMFd.vonMisesFisher_pdf_log(Xgrid[:,i],[mu,kappa]) )) probs = np.array(probs) probs = probs.reshape((probs.size,1)).T # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) # print X1_w.shape, X2_w.shape gl.plot(X1_w,X2_w, alpha = 1, lw = 3, ls = "-.", legend = ["Kvmf(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) gl.set_zoom(xlim = [-6,6], ylim = [-6,6], ax = ax1) ax2 = gl.subplot2grid((1,2), (0,1), rowspan=1, colspan=1) if (indx == 0): gl.add_text(positionXY = [0.1,.5], text = r' Initilization Incomplete LogLike: %.2f'%(logl[0]),fontsize = 15) pass elif (indx >= 1): gl.plot(range(1,np.array(logl).flatten()[1:].size +1),np.array(logl).flatten()[1:(indx+1)], ax = ax2, legend = ["Iteration %i, Incom LL: %.2f"%(indx, logl[indx])], labels = ["Convergence of LL with generated data","Iterations","LL"], lw = 2) gl.scatter(1, logl[1], lw = 2) pt = 0.05 gl.set_zoom(xlim = [0,len(logl)], ylim = [logl[1] - (logl[-1]-logl[1])*pt,logl[-1] + (logl[-1]-logl[1])*pt], ax = ax2) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.2, hspace=0.01) gl.savefig(folder_images_gif +'gif_'+ str(indx) + '.png', dpi = 100, sizeInches = [16, 8], close = "yes",bbox_inches = None) gl.close("all")
def generate_gaussian_data(folder_images, plot_original_data, N1 = 200, N2 = 300, N3 = 50): mu1 = np.array([[0],[0]]) cov1 = np.array([[0.8,-1.1], [-1.1,1.6]]) mu2 = np.array([[0],[0]]) cov2 = np.array([[0.3,0.45], [0.45,0.8]]) mu3 = np.array([[0],[0]]) cov3 = np.array([[0.1,0.0], [0.0,0.1]]) X1 = np.random.multivariate_normal(mu1.flatten(), cov1, N1).T X2 = np.random.multivariate_normal(mu2.flatten(), cov2, N2).T X3 = np.random.multivariate_normal(mu3.flatten(), cov3, N3).T # samples_X1 = np.array(range(X1.shape[1]))[np.where([X1[0,:] > 0])[0]] # samples_X1 = np.where(X1[0,:] > 0)[0] # np.array(range(X1.shape[1])) # print samples_X1 # X1 = X1[:,samples_X1] # X2 = np.concatenate((X2,X3),axis = 1) ######## Plotting ##### if (plot_original_data): gl.init_figure(); ## First cluster ax1 = gl.scatter(X1[0,:],X1[1,:], labels = ["Gaussian Generated Data", "x1","x2"], legend = ["K = 1"], color = "r",alpha = 0.5) mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = mu1, Sigma = cov1, Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "--", lw = 2 ,AxesStyle = "Normal2", color = "r") ## Second cluster ax1 = gl.scatter(X2[0,:],X2[1,:], legend = ["K = 2"], color = "b", alpha = 0.5) mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = mu2, Sigma = cov2, Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "--", lw = 2,AxesStyle = "Normal2", color = "b") ## Third cluster ax1 = gl.scatter(X3[0,:],X3[1,:], legend = ["K = 3"], color = "g", alpha = 0.5) mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = mu3, Sigma = cov3, Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "--", lw = 2,AxesStyle = "Normal2", color = "g") ax1.axis('equal') gl.savefig(folder_images +'Original data.png', dpi = 100, sizeInches = [12, 6]) ############ ESTIMATE THEM ################ theta1 = Gae.get_Gaussian_muSigma_ML(X1.T, parameters = dict([["Sigma","full"]])) print ("mu1:") print (theta1[0]) print ("Sigma1") print(theta1[1]) ############## Estimate Likelihood ################### ll = Gad.Gaussian_pdf_log (X1, [mu1,cov1]) ll2 = [] for i in range (ll.size): ll2.append( multivariate_normal.logpdf(X1[:,i], mean=mu1.flatten(), cov=cov1)) ll2 = np.array(ll2).reshape(ll.shape) print ("ll ours") print (ll.T) print ("ll scipy") print (ll2.T) print ("Difference in ll") print ((ll - ll2).T) ###### Multiple clusters case ll_K = Gad.Gaussian_K_pdf_log(X1, [[mu1,cov1],[mu2,cov2]]) if(0): X1 = gf.remove_module(X1.T).T X2 = gf.remove_module(X2.T).T X3 = gf.remove_module(X3.T).T Xdata = np.concatenate((X1,X2,X3), axis =1).T return X1,X2,X3,Xdata, mu1,mu2,mu3, cov1,cov2, cov3
def plot_final_distribution(Xs,mus,covs, Ks ,myDManager, logl,theta_list,model_theta_list, folder_images): colors = ["r","b","g"] K_G,K_W,K_vMF = Ks ################## Print the Watson and Gaussian Distribution parameters ################### for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): print ("------------ Gaussian Cluster. K = %i--------------------"%k) print ("mu") print (theta_list[-1][k][0]) print ("Sigma") print (theta_list[-1][k][1]) elif(distribution_name == "Watson"): print ("------------ Watson Cluster. K = %i--------------------"%k) print ("mu") print (theta_list[-1][k][0]) print ("Kappa") print (theta_list[-1][k][1]) elif(distribution_name == "vonMisesFisher"): print ("------------ vonMisesFisher Cluster. K = %i--------------------"%k) print ("mu") print (theta_list[-1][k][0]) print ("Kappa") print (theta_list[-1][k][1]) print ("pimix") print (model_theta_list[-1]) mus_Watson_Gaussian = [] # k_c is the number of the cluster inside the Manager. k is the index in theta for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W mus_k = [] for iter_i in range(len(theta_list)): # For each iteration of the algorihtm if (distribution_name == "Gaussian"): theta_i = theta_list[iter_i][k] mus_k.append(theta_i[0]) elif(distribution_name == "Watson"): theta_i = theta_list[iter_i][k] mus_k.append(theta_i[0]) elif(distribution_name == "vonMisesFisher"): theta_i = theta_list[iter_i][k] mus_k.append(theta_i[0]) mus_k = np.concatenate(mus_k, axis = 1).T mus_Watson_Gaussian.append(mus_k) ######## Plot the original data ##### gl.init_figure(); ## First cluster for xi in range(len( Xs)): ## First cluster ax1 = gl.scatter(Xs[xi][0,:],Xs[xi][1,:], labels = ['EM Evolution. Kg:'+str(K_G)+ ', Kw:' + str(K_W) + ', K_vMF:' + str(K_vMF), "X1","X2"], color = colors[xi] ,alpha = 0.2, nf = 0) mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = mus[xi], Sigma = covs[xi], Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "--", lw = 2 ,AxesStyle = "Normal2", color = colors[xi], alpha = 0.7) indx = -1 # Only doable if the clusters dont die Nit,Ndim = mus_Watson_Gaussian[0].shape for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): ## Plot the ecolution of the mu #### Plot the Covariance of the clusters ! mean,w,h,theta = bMA.get_gaussian_ellipse_params( mu = theta_list[indx][k][0], Sigma = theta_list[indx][k][1], Chi2val = 2.4477) r_ellipse = bMA.get_ellipse_points(mean,w,h,theta) gl.plot(r_ellipse[:,0], r_ellipse[:,1], ax = ax1, ls = "-.", lw = 3, AxesStyle = "Normal2", legend = ["Kg(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) gl.scatter(mus_Watson_Gaussian[k][:,0], mus_Watson_Gaussian[k][:,1], nf = 0, na = 0, alpha = 0.3, lw = 1, color = "y") gl.plot(mus_Watson_Gaussian[k][:,0], mus_Watson_Gaussian[k][:,1], nf = 0, na = 0, alpha = 0.8, lw = 2, color = "y") elif(distribution_name == "Watson"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2*np.pi, Nsa) Xgrid= np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append(np.exp(Wad.Watson_pdf_log(Xgrid[:,i],[mu,kappa]) )) probs = np.array(probs) # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) gl.plot(X1_w,X2_w, legend = ["Kw(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))] , alpha = 1, lw = 3, ls = "-.") elif(distribution_name == "vonMisesFisher"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2*np.pi, Nsa) Xgrid= np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append(np.exp(vMFd.vonMisesFisher_pdf_log(Xgrid[:,i],[mu,kappa]) )) probs = np.array(probs) probs = probs.reshape((probs.size,1)).T # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) # print X1_w.shape, X2_w.shape gl.plot(X1_w,X2_w, alpha = 1, lw = 3, ls = "-.", legend = ["Kvmf(%i). pi:%0.2f"%(k, float(model_theta_list[indx][0][0,k]))]) ax1.axis('equal') gl.savefig(folder_images +'Final_State. K_G:'+str(K_G)+ ', K_W:' + str(K_W) + ', K_vMF:' + str(K_vMF) + '.png', dpi = 100, sizeInches = [12, 6])
def create_Bayesian_analysis_charts(model, X_data_tr, Y_data_tr, X_data_val, Y_data_val, tr_loss, val_loss, KL_loss,final_loss_tr,final_loss_val, xgrid_real_func, ygrid_real_func, folder_images, epoch_i = None): # Configurations of the plots alpha_points = 0.2 color_points_train = "dark navy blue" color_points_val = "amber" color_train_loss = "cobalt blue" color_val_loss = "blood" color_truth = "k" color_mean = "b" color_most_likey = "y" ############################# Data computation ####################### if(type(X_data_tr) == type([])): pass else: if (X_data_tr.shape[1] == 1): # Regression Example x_grid, all_y_grid,most_likely_ygrid = compute_regression_1D_data( model,X_data_tr,X_data_val, Nsamples = 100) elif(X_data_tr.shape[1] == 2): # Classification Example xx,yy , all_y_grid,most_likely_ygrid = compute_classification_2D_data( model,X_data_tr,X_data_val, Nsamples = 100) else: # RNN x_grid, all_y_grid,most_likely_ygrid = compute_RNN_1D_data( model,X_data_tr,X_data_val, Nsamples = 100) ################################ Divide in plots ############################## gl.init_figure(); ax1 = gl.subplot2grid((6,3), (0,0), rowspan=3, colspan=1) ax2 = gl.subplot2grid((6,3), (3,0), rowspan=3, colspan=1, sharex = ax1, sharey = ax1) ax3 = gl.subplot2grid((6,3), (0,1), rowspan=2, colspan=1) ax4 = gl.subplot2grid((6,3), (2,1), rowspan=2, colspan=1, sharex = ax3) ax5 = gl.subplot2grid((6,3), (4,1), rowspan=2, colspan=1, sharex = ax3) ax6 = gl.subplot2grid((6,3), (0,2), rowspan=3, colspan=1) ax7 = gl.subplot2grid((6,3), (3,2), rowspan=3, colspan=1, sharex = ax6) if(type(X_data_tr) == type([])): Xtrain = [torch.tensor(X_data_tr[i],device=model.cf_a.device, dtype=model.cf_a.dtype) for i in range(len(X_data_tr))] Ytrain = torch.tensor(Y_data_tr,device=model.cf_a.device, dtype=torch.int64) Xval = [torch.tensor(X_data_val[i],device=model.cf_a.device, dtype=model.cf_a.dtype) for i in range(len(X_data_val))] Yval = torch.tensor(Y_data_val,device=model.cf_a.device, dtype=torch.int64) confusion = model.get_confusion_matrix(Xtrain, Ytrain) plot_confusion_matrix(confusion,model.languages, ax1 ) confusion = model.get_confusion_matrix(Xval, Yval) plot_confusion_matrix(confusion,model.languages, ax2 ) else: if (X_data_tr.shape[1] == 1): # Regression Example plot_data_regression_1d_2axes(X_data_tr, Y_data_tr, xgrid_real_func, ygrid_real_func, X_data_val, Y_data_val, x_grid,all_y_grid, most_likely_ygrid, alpha_points, color_points_train, color_points_val, color_most_likey,color_mean,color_truth, ax1,ax2) elif(X_data_tr.shape[1] == 2): # Classification Example plot_data_classification_2d_2axes(X_data_tr, Y_data_tr, xgrid_real_func, ygrid_real_func, X_data_val, Y_data_val, xx,yy,all_y_grid, most_likely_ygrid, alpha_points, color_points_train, color_points_val, color_most_likey,color_mean, color_truth, ax1,ax2) else: # RNN example plot_data_RNN_1d_2axes(X_data_tr, Y_data_tr, xgrid_real_func, ygrid_real_func, X_data_val, Y_data_val, x_grid,all_y_grid, most_likely_ygrid, alpha_points, color_points_train, color_points_val, color_most_likey,color_mean,color_truth, ax1,ax2) # gl.fill_between (x_grid, [mean_samples_grid + 2*std_samples_grid, mean_samples_grid - 2*std_samples_grid] # , ax = ax2, alpha = 0.10, color = "b", legend = ["Mean realizaions"]) ## ax2: The uncertainty of the prediction !! # gl.plot (x_grid, std_samples_grid, ax = ax2, labels = ["Std (%i)"%(Nsamples),"X","f(X)"], legend = [" std predictions"], fill = 1, alpha = 0.3) ############## ax3 ax4 ax5: Loss Evolution !! ###################### ## ax3: Evolutoin of the data loss gl.plot([], tr_loss, ax = ax3, lw = 3, labels = ["Losses", "","Data loss"], legend = ["train"], color = color_train_loss) gl.plot([], val_loss,ax = ax3, lw = 3, legend = ["validation"], color = color_val_loss, AxesStyle = "Normal - No xaxis") ## ax4: The evolution of the KL loss gl.plot([], KL_loss, ax = ax4, lw = 3, labels = ["", "","KL loss"], legend = ["Bayesian Weights"], AxesStyle = "Normal - No xaxis", color = "k") ## ax5: Evolutoin of the total loss gl.plot([], final_loss_tr, ax = ax5, lw = 3, labels = ["", "epoch","Total Loss (Bayes)"], legend = ["train"], color = color_train_loss) gl.plot([], final_loss_val,ax = ax5, lw = 3, legend = ["validation"], color = color_val_loss) ############## ax6 ax7: Variational Weights !! ###################### create_plot_variational_weights(model,ax6,ax7) ## Plot in chart 7 the acceptable mu = 2sigma -> sigma = |mu|/2sigma mu_grid = np.linspace(-3,3,100) y_grid = np.abs(mu_grid)/2 gl.fill_between(mu_grid, 10*np.ones(mu_grid.size), y_grid, alpha = 0.2, color = "r", ax = ax7, legend = ["95% non-significant"]) gl.set_zoom (ax = ax6, ylim = [-0.1,10]) gl.set_zoom (ax = ax7, xlim = [-2.5, 2.5], ylim = [-0.05, np.exp(model.cf_a.input_layer_prior["log_sigma2"])*(1 + 0.15)]) # gl.set_zoom (ax = ax7, xlim = [-2.5, 2.5], ylim = [-0.1,2]) # Set final properties and save figure gl.set_fontSizes(ax = [ax1,ax2,ax3,ax4,ax5,ax6,ax7], title = 20, xlabel = 20, ylabel = 20, legend = 10, xticks = 12, yticks = 12) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.10) if (type(epoch_i) == type(None)): gl.savefig(folder_images +"../"+'Final_values_regression_1D_' +str(model.cf_a.eta_KL) +'.png', dpi = 100, sizeInches = [20, 10]) else: gl.savefig(folder_images +'%i.png'%epoch_i, dpi = 100, sizeInches = [20, 10], close = True, bbox_inches = "tight")
def create_Bayesian_analysis_charts_simplified(model, train_dataset, validation_dataset, tr_loss, val_loss, KL_loss, folder_images, epoch_i = None): # Configurations of the plots alpha_points = 0.2 color_points_train = "dark navy blue" color_points_val = "amber" color_train_loss = "cobalt blue" color_val_loss = "blood" color_truth = "k" color_mean = "b" color_most_likey = "y" ################################ Divide in plots ############################## gl.init_figure(); ax1 = gl.subplot2grid((6,3), (0,0), rowspan=3, colspan=1) ax2 = gl.subplot2grid((6,3), (3,0), rowspan=3, colspan=1, sharex = ax1, sharey = ax1) ax3 = gl.subplot2grid((6,3), (0,1), rowspan=2, colspan=1) ax4 = gl.subplot2grid((6,3), (2,1), rowspan=2, colspan=1, sharex = ax3) ax5 = gl.subplot2grid((6,3), (4,1), rowspan=2, colspan=1, sharex = ax3) ax6 = gl.subplot2grid((6,3), (0,2), rowspan=3, colspan=1) ax7 = gl.subplot2grid((6,3), (3,2), rowspan=3, colspan=1, sharex = ax6) ####### ax1, ax2: Get confusion matrices ########## labels_classes, confusion = model.get_confusion_matrix(train_dataset) plot_confusion_matrix(confusion,labels_classes, ax1 ) labels_classes, confusion = model.get_confusion_matrix(validation_dataset) plot_confusion_matrix(confusion,labels_classes, ax2 ) ############## ax3 ax4 ax5: Loss Evolution !! ###################### ## ax3: Evolutoin of the data loss gl.plot([], tr_loss, ax = ax3, lw = 3, labels = ["Losses", "","Data loss (MSE)"], legend = ["train"], color = color_train_loss) gl.plot([], val_loss,ax = ax3, lw = 3, legend = ["validation"], color = color_val_loss, AxesStyle = "Normal - No xaxis") ## ax4: The evolution of the KL loss gl.plot([], KL_loss, ax = ax4, lw = 3, labels = ["", "","KL loss"], legend = ["Bayesian Weights"], AxesStyle = "Normal - No xaxis", color = "k") ## ax5: Evolutoin of the total loss gl.plot([], tr_loss, ax = ax5, lw = 3, labels = ["", "epoch","Total Loss (Bayes)"], legend = ["train"], color = color_train_loss) gl.plot([], val_loss,ax = ax5, lw = 3, legend = ["validation"], color = color_val_loss) ############## ax6 ax7: Variational Weights !! ###################### create_plot_variational_weights(model,ax6,ax7) gl.set_zoom (ax = ax6, ylim = [-0.1,10]) gl.set_zoom (ax = ax7, xlim = [-2.5, 2.5], ylim = [-0.1,0.5]) # Set final properties and save figure gl.set_fontSizes(ax = [ax1,ax2,ax3,ax4,ax5,ax6,ax7], title = 20, xlabel = 20, ylabel = 20, legend = 10, xticks = 12, yticks = 12) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.10) if (type(epoch_i) == type(None)): gl.savefig(folder_images +'Training_Example_Data_Bayesian.png', dpi = 100, sizeInches = [20, 10]) else: gl.savefig(folder_images +'%i.png'%epoch_i, dpi = 100, sizeInches = [20, 10], close = True, bbox_inches = "tight")
def create_Bayesian_analysis_charts(model, X_data_tr, Y_data_tr, X_data_val, Y_data_val, tr_loss, val_loss, KL_loss, final_loss_tr, final_loss_val, xgrid_real_func, ygrid_real_func, folder_images, epoch_i=None): # Configurations of the plots alpha_points = 0.2 color_points_train = "dark navy blue" color_points_val = "amber" color_train_loss = "cobalt blue" color_val_loss = "blood" color_truth = "k" color_mean = "b" color_most_likey = "y" ############################# Data computation ####################### if (type(X_data_tr) == type([])): pass else: if (X_data_tr.shape[1] == 1): # Regression Example x_grid, all_y_grid, most_likely_ygrid = compute_regression_1D_data( model, X_data_tr, X_data_val, Nsamples=100) elif (X_data_tr.shape[1] == 2): # Classification Example xx, yy, all_y_grid, most_likely_ygrid = compute_classification_2D_data( model, X_data_tr, X_data_val, Nsamples=100) else: # RNN x_grid, all_y_grid, most_likely_ygrid = compute_RNN_1D_data( model, X_data_tr, X_data_val, Nsamples=100) ################################ Divide in plots ############################## gl.init_figure() ax1 = gl.subplot2grid((6, 3), (0, 0), rowspan=3, colspan=1) ax2 = gl.subplot2grid((6, 3), (3, 0), rowspan=3, colspan=1, sharex=ax1, sharey=ax1) ax3 = gl.subplot2grid((6, 3), (0, 1), rowspan=2, colspan=1) ax4 = gl.subplot2grid((6, 3), (2, 1), rowspan=2, colspan=1, sharex=ax3) ax5 = gl.subplot2grid((6, 3), (4, 1), rowspan=2, colspan=1, sharex=ax3) ax6 = gl.subplot2grid((6, 3), (0, 2), rowspan=3, colspan=1) ax7 = gl.subplot2grid((6, 3), (3, 2), rowspan=3, colspan=1, sharex=ax6) if (type(X_data_tr) == type([])): Xtrain = [ torch.tensor(X_data_tr[i], device=model.cf_a.device, dtype=model.cf_a.dtype) for i in range(len(X_data_tr)) ] Ytrain = torch.tensor(Y_data_tr, device=model.cf_a.device, dtype=torch.int64) Xval = [ torch.tensor(X_data_val[i], device=model.cf_a.device, dtype=model.cf_a.dtype) for i in range(len(X_data_val)) ] Yval = torch.tensor(Y_data_val, device=model.cf_a.device, dtype=torch.int64) confusion = model.get_confusion_matrix(Xtrain, Ytrain) plot_confusion_matrix(confusion, model.languages, ax1) confusion = model.get_confusion_matrix(Xval, Yval) plot_confusion_matrix(confusion, model.languages, ax2) else: if (X_data_tr.shape[1] == 1): # Regression Example plot_data_regression_1d_2axes( X_data_tr, Y_data_tr, xgrid_real_func, ygrid_real_func, X_data_val, Y_data_val, x_grid, all_y_grid, most_likely_ygrid, alpha_points, color_points_train, color_points_val, color_most_likey, color_mean, color_truth, ax1, ax2) elif (X_data_tr.shape[1] == 2): # Classification Example plot_data_classification_2d_2axes( X_data_tr, Y_data_tr, xgrid_real_func, ygrid_real_func, X_data_val, Y_data_val, xx, yy, all_y_grid, most_likely_ygrid, alpha_points, color_points_train, color_points_val, color_most_likey, color_mean, color_truth, ax1, ax2) else: # RNN example plot_data_RNN_1d_2axes(X_data_tr, Y_data_tr, xgrid_real_func, ygrid_real_func, X_data_val, Y_data_val, x_grid, all_y_grid, most_likely_ygrid, alpha_points, color_points_train, color_points_val, color_most_likey, color_mean, color_truth, ax1, ax2) # gl.fill_between (x_grid, [mean_samples_grid + 2*std_samples_grid, mean_samples_grid - 2*std_samples_grid] # , ax = ax2, alpha = 0.10, color = "b", legend = ["Mean realizaions"]) ## ax2: The uncertainty of the prediction !! # gl.plot (x_grid, std_samples_grid, ax = ax2, labels = ["Std (%i)"%(Nsamples),"X","f(X)"], legend = [" std predictions"], fill = 1, alpha = 0.3) ############## ax3 ax4 ax5: Loss Evolution !! ###################### ## ax3: Evolutoin of the data loss gl.plot([], tr_loss, ax=ax3, lw=3, labels=["Losses", "", "Data loss"], legend=["train"], color=color_train_loss) gl.plot([], val_loss, ax=ax3, lw=3, legend=["validation"], color=color_val_loss, AxesStyle="Normal - No xaxis") ## ax4: The evolution of the KL loss gl.plot([], KL_loss, ax=ax4, lw=3, labels=["", "", "KL loss"], legend=["Bayesian Weights"], AxesStyle="Normal - No xaxis", color="k") ## ax5: Evolutoin of the total loss gl.plot([], final_loss_tr, ax=ax5, lw=3, labels=["", "epoch", "Total Loss (Bayes)"], legend=["train"], color=color_train_loss) gl.plot([], final_loss_val, ax=ax5, lw=3, legend=["validation"], color=color_val_loss) ############## ax6 ax7: Variational Weights !! ###################### create_plot_variational_weights(model, ax6, ax7) ## Plot in chart 7 the acceptable mu = 2sigma -> sigma = |mu|/2sigma mu_grid = np.linspace(-3, 3, 100) y_grid = np.abs(mu_grid) / 2 gl.fill_between(mu_grid, 10 * np.ones(mu_grid.size), y_grid, alpha=0.2, color="r", ax=ax7, legend=["95% non-significant"]) gl.set_zoom(ax=ax6, ylim=[-0.1, 10]) gl.set_zoom(ax=ax7, xlim=[-2.5, 2.5], ylim=[ -0.05, np.exp(model.cf_a.input_layer_prior["log_sigma2"]) * (1 + 0.15) ]) # gl.set_zoom (ax = ax7, xlim = [-2.5, 2.5], ylim = [-0.1,2]) # Set final properties and save figure gl.set_fontSizes(ax=[ax1, ax2, ax3, ax4, ax5, ax6, ax7], title=20, xlabel=20, ylabel=20, legend=10, xticks=12, yticks=12) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.10) if (type(epoch_i) == type(None)): gl.savefig(folder_images + "../" + 'Final_values_regression_1D_' + str(model.cf_a.eta_KL) + '.png', dpi=100, sizeInches=[20, 10]) else: gl.savefig(folder_images + '%i.png' % epoch_i, dpi=100, sizeInches=[20, 10], close=True, bbox_inches="tight")
def visualize_attention_matrix(question_tokens, passage_tokens, attention_matrix, image_path): """ Text to visualze attention map for.a given exmaple. question_tokens: List of tokens of the question passage_tokens: List of tokens of the passage attention_matrix: len(passage) x len(question) matrix with the probabilities """ f = gl.init_figure() ax = f.add_axes([0.1, 0.3, 0.8, 0.5]) ax_attention_words = f.add_axes([0.1, 0.70, 0.8, 0.15]) ax_attention_words.axis('off') # add image cmap = "binary" #cm.get_cmap('coolwarm', 30) i = ax.imshow(attention_matrix, interpolation='nearest', cmap=cmap,vmin=0, vmax=1) # add colorbar cbaxes = f.add_axes([0.95, 0.3, 0.02, 0.5]) cbar = f.colorbar(i, cax=cbaxes, orientation='vertical') cbar.ax.set_xlabel('Probability', labelpad=6) # add labels ax.set_yticks(range(len(question_tokens))) ax.set_yticklabels(question_tokens) ax.set_xticks(range(len(passage_tokens))) ax.set_xticklabels(passage_tokens, rotation=80) ax.set_xlabel('Passage') ax.set_ylabel('Question') ########### GET THE MOST ATTENTION WORDS ######## Nmax_attention_words = 3 z = (-attention_matrix).argsort(axis = 1)[:,:] attentioned_passage_words = [] for i in range (len(question_tokens)): attentioned_passage_words.append([]) for j in range(Nmax_attention_words): attentioned_passage_words[-1].append(passage_tokens[z[i,j]] + "(%.1f%%)"%(attention_matrix[i,z[i,j]]*100)) attentioned_passage_words[-1] = ", ".join(attentioned_passage_words[-1]) text_correspondance = "" for i in range (len(question_tokens)): text_correspondance += question_tokens[i] + " ---> " + attentioned_passage_words[i] + "\n" ax_attention_words.text(0,0,text_correspondance) # ax2.yaxis.tick_right() # ax2.yaxis.set_label_position("right") f.show() # gl.set_fontSizes(ax = [ax1,ax2], title = 20, xlabel = 15, ylabel = 18, # legend = 12, xticks = 14, yticks = 14) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.20, hspace=0.10) gl.savefig(image_path, dpi = 100, sizeInches = [10, 6], close = False, bbox_inches = "tight")
def generate_images_iterations_ll(Xs, mus, covs, Ks, myDManager, logl, theta_list, model_theta_list, folder_images_gif): # os.remove(folder_images_gif) # Remove previous images if existing """ WARNING: MEANT FOR ONLY 3 Distributions due to the color RGB """ import shutil ul.create_folder_if_needed(folder_images_gif) shutil.rmtree(folder_images_gif) ul.create_folder_if_needed(folder_images_gif) ######## Plot the original data ##### Xdata = np.concatenate(Xs, axis=1).T colors = ["r", "b", "g"] K_G, K_W, K_vMF = Ks ### FOR EACH ITERATION for i in range(len(theta_list)): # theta_list indx = i gl.init_figure() ax1 = gl.subplot2grid((1, 2), (0, 0), rowspan=1, colspan=1) ## Get the relative ll of the Gaussian denoising cluster. ll = myDManager.pdf_log_K(Xdata, theta_list[indx]) N, K = ll.shape # print ll.shape for j in range(N): # For every sample #TODO: Can this not be done without a for ? # Normalize the probability of the sample being generated by the clusters Marginal_xi_probability = gf.sum_logs(ll[j, :]) ll[j, :] = ll[j, :] - Marginal_xi_probability ax1 = gl.scatter( Xdata[j, 0], Xdata[j, 1], labels=[ 'EM Evolution. Kg:' + str(K_G) + ', Kw:' + str(K_W) + ', K_vMF:' + str(K_vMF), "X1", "X2" ], color=(np.exp(ll[j, 1]), np.exp(ll[j, 0]), np.exp(ll[j, 2])), ### np.exp(ll[j,2]) alpha=1, nf=0) # Only doable if the clusters dont die for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): ## Plot the ecolution of the mu #### Plot the Covariance of the clusters ! mean, w, h, theta = bMA.get_gaussian_ellipse_params( mu=theta_list[indx][k][0], Sigma=theta_list[indx][k][1], Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="-.", lw=3, AxesStyle="Normal2", legend=[ "Kg(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) elif (distribution_name == "Watson"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[-1][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2 * np.pi, Nsa) Xgrid = np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append( np.exp(Wad.Watson_pdf_log(Xgrid[:, i], [mu, kappa]))) probs = np.array(probs) # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) gl.plot(X1_w, X2_w, alpha=1, lw=3, ls="-.", legend=[ "Kw(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) elif (distribution_name == "vonMisesFisher"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2 * np.pi, Nsa) Xgrid = np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append( np.exp( vMFd.vonMisesFisher_pdf_log( Xgrid[:, i], [mu, kappa]))) probs = np.array(probs) probs = probs.reshape((probs.size, 1)).T # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) # print X1_w.shape, X2_w.shape gl.plot(X1_w, X2_w, alpha=1, lw=3, ls="-.", legend=[ "Kvmf(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) gl.set_zoom(xlim=[-6, 6], ylim=[-6, 6], ax=ax1) ax2 = gl.subplot2grid((1, 2), (0, 1), rowspan=1, colspan=1) if (indx == 0): gl.add_text(positionXY=[0.1, .5], text=r' Initilization Incomplete LogLike: %.2f' % (logl[0]), fontsize=15) pass elif (indx >= 1): gl.plot( range(1, np.array(logl).flatten()[1:].size + 1), np.array(logl).flatten()[1:(indx + 1)], ax=ax2, legend=["Iteration %i, Incom LL: %.2f" % (indx, logl[indx])], labels=[ "Convergence of LL with generated data", "Iterations", "LL" ], lw=2) gl.scatter(1, logl[1], lw=2) pt = 0.05 gl.set_zoom(xlim=[0, len(logl)], ylim=[ logl[1] - (logl[-1] - logl[1]) * pt, logl[-1] + (logl[-1] - logl[1]) * pt ], ax=ax2) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.2, hspace=0.01) gl.savefig(folder_images_gif + 'gif_' + str(indx) + '.png', dpi=100, sizeInches=[16, 8], close="yes", bbox_inches=None) gl.close("all")
def plot_multiple_iterations(Xs, mus, covs, Ks, myDManager, logl, theta_list, model_theta_list, folder_images): ######## Plot the original data ##### gl.init_figure() gl.set_subplots(2, 3) Ngraph = 6 colors = ["r", "b", "g"] K_G, K_W, K_vMF = Ks for i in range(Ngraph): indx = int(i * ((len(theta_list) - 1) / float(Ngraph - 1))) nf = 1 for xi in range(len(Xs)): ## First cluster labels = [ 'EM Evolution. Kg:' + str(K_G) + ', Kw:' + str(K_W) + ', K_vMF:' + str(K_vMF), "X1", "X2" ] ax1 = gl.scatter(Xs[xi][0, :], Xs[xi][1, :], labels=["", "", ""], color=colors[xi], alpha=0.2, nf=nf) nf = 0 mean, w, h, theta = bMA.get_gaussian_ellipse_params(mu=mus[xi], Sigma=covs[xi], Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="--", lw=2, AxesStyle="Normal2", color=colors[xi], alpha=0.7) # Only doable if the clusters dont die for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): ## Plot the ecolution of the mu #### Plot the Covariance of the clusters ! mean, w, h, theta = bMA.get_gaussian_ellipse_params( mu=theta_list[indx][k][0], Sigma=theta_list[indx][k][1], Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="-.", lw=3, AxesStyle="Normal2", legend=[ "Kg(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) elif (distribution_name == "Watson"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2 * np.pi, Nsa) Xgrid = np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append( np.exp(Wad.Watson_pdf_log(Xgrid[:, i], [mu, kappa]))) probs = np.array(probs) # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) gl.plot(X1_w, X2_w, alpha=1, lw=3, ls="-.", legend=[ "Kw(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) elif (distribution_name == "vonMisesFisher"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2 * np.pi, Nsa) Xgrid = np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append( np.exp( vMFd.vonMisesFisher_pdf_log( Xgrid[:, i], [mu, kappa]))) probs = np.array(probs) probs = probs.reshape((probs.size, 1)).T # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) # print X1_w.shape, X2_w.shape gl.plot(X1_w, X2_w, alpha=1, lw=3, ls="-.", legend=[ "Kvmf(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) ax1.axis('equal') gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.2, hspace=0.01) gl.savefig(folder_images + 'Final_State2. K_G:' + str(K_G) + ', K_W:' + str(K_W) + '.png', dpi=100, sizeInches=[18, 8])
def plot_final_distribution(Xs, mus, covs, Ks, myDManager, logl, theta_list, model_theta_list, folder_images): colors = ["r", "b", "g"] K_G, K_W, K_vMF = Ks ################## Print the Watson and Gaussian Distribution parameters ################### for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): print("------------ Gaussian Cluster. K = %i--------------------" % k) print("mu") print(theta_list[-1][k][0]) print("Sigma") print(theta_list[-1][k][1]) elif (distribution_name == "Watson"): print("------------ Watson Cluster. K = %i--------------------" % k) print("mu") print(theta_list[-1][k][0]) print("Kappa") print(theta_list[-1][k][1]) elif (distribution_name == "vonMisesFisher"): print( "------------ vonMisesFisher Cluster. K = %i--------------------" % k) print("mu") print(theta_list[-1][k][0]) print("Kappa") print(theta_list[-1][k][1]) print("pimix") print(model_theta_list[-1]) mus_Watson_Gaussian = [] # k_c is the number of the cluster inside the Manager. k is the index in theta for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W mus_k = [] for iter_i in range( len(theta_list)): # For each iteration of the algorihtm if (distribution_name == "Gaussian"): theta_i = theta_list[iter_i][k] mus_k.append(theta_i[0]) elif (distribution_name == "Watson"): theta_i = theta_list[iter_i][k] mus_k.append(theta_i[0]) elif (distribution_name == "vonMisesFisher"): theta_i = theta_list[iter_i][k] mus_k.append(theta_i[0]) mus_k = np.concatenate(mus_k, axis=1).T mus_Watson_Gaussian.append(mus_k) ######## Plot the original data ##### gl.init_figure() ## First cluster for xi in range(len(Xs)): ## First cluster ax1 = gl.scatter(Xs[xi][0, :], Xs[xi][1, :], labels=[ 'EM Evolution. Kg:' + str(K_G) + ', Kw:' + str(K_W) + ', K_vMF:' + str(K_vMF), "X1", "X2" ], color=colors[xi], alpha=0.2, nf=0) mean, w, h, theta = bMA.get_gaussian_ellipse_params(mu=mus[xi], Sigma=covs[xi], Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="--", lw=2, AxesStyle="Normal2", color=colors[xi], alpha=0.7) indx = -1 # Only doable if the clusters dont die Nit, Ndim = mus_Watson_Gaussian[0].shape for k_c in myDManager.clusterk_to_Dname.keys(): k = myDManager.clusterk_to_thetak[k_c] distribution_name = myDManager.clusterk_to_Dname[k_c] # G W if (distribution_name == "Gaussian"): ## Plot the ecolution of the mu #### Plot the Covariance of the clusters ! mean, w, h, theta = bMA.get_gaussian_ellipse_params( mu=theta_list[indx][k][0], Sigma=theta_list[indx][k][1], Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="-.", lw=3, AxesStyle="Normal2", legend=[ "Kg(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) gl.scatter(mus_Watson_Gaussian[k][:, 0], mus_Watson_Gaussian[k][:, 1], nf=0, na=0, alpha=0.3, lw=1, color="y") gl.plot(mus_Watson_Gaussian[k][:, 0], mus_Watson_Gaussian[k][:, 1], nf=0, na=0, alpha=0.8, lw=2, color="y") elif (distribution_name == "Watson"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2 * np.pi, Nsa) Xgrid = np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append( np.exp(Wad.Watson_pdf_log(Xgrid[:, i], [mu, kappa]))) probs = np.array(probs) # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) gl.plot(X1_w, X2_w, legend=[ "Kw(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ], alpha=1, lw=3, ls="-.") elif (distribution_name == "vonMisesFisher"): #### Plot the pdf of the distributino ! ## Distribution parameters for Watson kappa = float(theta_list[indx][k][1]) mu = theta_list[indx][k][0] Nsa = 1000 # Draw 2D samples as transformation of the angle Xalpha = np.linspace(0, 2 * np.pi, Nsa) Xgrid = np.array([np.cos(Xalpha), np.sin(Xalpha)]) probs = [] # Vector with probabilities for i in range(Nsa): probs.append( np.exp( vMFd.vonMisesFisher_pdf_log(Xgrid[:, i], [mu, kappa]))) probs = np.array(probs) probs = probs.reshape((probs.size, 1)).T # Plot it in polar coordinates X1_w = (1 + probs) * np.cos(Xalpha) X2_w = (1 + probs) * np.sin(Xalpha) # print X1_w.shape, X2_w.shape gl.plot(X1_w, X2_w, alpha=1, lw=3, ls="-.", legend=[ "Kvmf(%i). pi:%0.2f" % (k, float(model_theta_list[indx][0][0, k])) ]) ax1.axis('equal') gl.savefig(folder_images + 'Final_State. K_G:' + str(K_G) + ', K_W:' + str(K_W) + ', K_vMF:' + str(K_vMF) + '.png', dpi=100, sizeInches=[12, 6])
def generate_gaussian_data(folder_images, plot_original_data, N1=200, N2=300, N3=50): mu1 = np.array([[0], [0]]) cov1 = np.array([[0.8, -1.1], [-1.1, 1.6]]) mu2 = np.array([[0], [0]]) cov2 = np.array([[0.3, 0.45], [0.45, 0.8]]) mu3 = np.array([[0], [0]]) cov3 = np.array([[0.1, 0.0], [0.0, 0.1]]) X1 = np.random.multivariate_normal(mu1.flatten(), cov1, N1).T X2 = np.random.multivariate_normal(mu2.flatten(), cov2, N2).T X3 = np.random.multivariate_normal(mu3.flatten(), cov3, N3).T # samples_X1 = np.array(range(X1.shape[1]))[np.where([X1[0,:] > 0])[0]] # samples_X1 = np.where(X1[0,:] > 0)[0] # np.array(range(X1.shape[1])) # print samples_X1 # X1 = X1[:,samples_X1] # X2 = np.concatenate((X2,X3),axis = 1) ######## Plotting ##### if (plot_original_data): gl.init_figure() ## First cluster ax1 = gl.scatter(X1[0, :], X1[1, :], labels=["Gaussian Generated Data", "x1", "x2"], legend=["K = 1"], color="r", alpha=0.5) mean, w, h, theta = bMA.get_gaussian_ellipse_params(mu=mu1, Sigma=cov1, Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="--", lw=2, AxesStyle="Normal2", color="r") ## Second cluster ax1 = gl.scatter(X2[0, :], X2[1, :], legend=["K = 2"], color="b", alpha=0.5) mean, w, h, theta = bMA.get_gaussian_ellipse_params(mu=mu2, Sigma=cov2, Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="--", lw=2, AxesStyle="Normal2", color="b") ## Third cluster ax1 = gl.scatter(X3[0, :], X3[1, :], legend=["K = 3"], color="g", alpha=0.5) mean, w, h, theta = bMA.get_gaussian_ellipse_params(mu=mu3, Sigma=cov3, Chi2val=2.4477) r_ellipse = bMA.get_ellipse_points(mean, w, h, theta) gl.plot(r_ellipse[:, 0], r_ellipse[:, 1], ax=ax1, ls="--", lw=2, AxesStyle="Normal2", color="g") ax1.axis('equal') gl.savefig(folder_images + 'Original data.png', dpi=100, sizeInches=[12, 6]) ############ ESTIMATE THEM ################ theta1 = Gae.get_Gaussian_muSigma_ML(X1.T, parameters=dict([["Sigma", "full"]])) print("mu1:") print(theta1[0]) print("Sigma1") print(theta1[1]) ############## Estimate Likelihood ################### ll = Gad.Gaussian_pdf_log(X1, [mu1, cov1]) ll2 = [] for i in range(ll.size): ll2.append( multivariate_normal.logpdf(X1[:, i], mean=mu1.flatten(), cov=cov1)) ll2 = np.array(ll2).reshape(ll.shape) print("ll ours") print(ll.T) print("ll scipy") print(ll2.T) print("Difference in ll") print((ll - ll2).T) ###### Multiple clusters case ll_K = Gad.Gaussian_K_pdf_log(X1, [[mu1, cov1], [mu2, cov2]]) if (0): X1 = gf.remove_module(X1.T).T X2 = gf.remove_module(X2.T).T X3 = gf.remove_module(X3.T).T Xdata = np.concatenate((X1, X2, X3), axis=1).T return X1, X2, X3, Xdata, mu1, mu2, mu3, cov1, cov2, cov3
def init_figure(self): """ This function initializes the chart, with its widgets and everything """ button_height = 0.030; textbox_length0 = 0.02 textbox_length1 = 0.04 textbox_length2 = 0.05 fig = gl.init_figure(); ## Set the image to full screen fig_manager = plt.get_current_fig_manager() if hasattr(fig_manager, 'window'): fig_manager.window.showMaximized() data_axes = gl.subplot2grid((1,4), (0,0), rowspan=1, colspan=3) self.fig = fig; self.data_axes = data_axes; #### Logo Images !! logo_path = self.output_folder + "images_IoTubes/IoTubes_logo.png" image = mpimg.imread(logo_path) ax_img = plt.axes([0.725, 0.75, 0.2, 0.2]) ax_img.imshow(image) ax_img.axis("off") ################## Widgets Axes ##################### widgets_x = 0.76 widgets_x2 = 0.85 widgets_x3 = 0.90 w1_x, w2_x, w3_x = 0.73, 0.8,0.87 base_y = 0.69 administration_y = base_y monitoring_y = administration_y - 0.12 chart_s_y = monitoring_y - 0.12 chart_s_y2 = chart_s_y -0.05 chart_start_stop_y = chart_s_y2 - 0.05 output_y = chart_start_stop_y - 0.12 diff_headline_content = 0.052 ## Administration ! headlines_x = 0.705 text = self.fig.text(headlines_x, administration_y + diff_headline_content, 'Administration:', size=20) # ha='center', va='center', size=20) axbox_machineID = plt.axes([widgets_x, administration_y, textbox_length1, button_height]) axbox_pipingID = plt.axes([widgets_x2, administration_y, textbox_length1, button_height]) ### Monitoring text = self.fig.text(headlines_x, monitoring_y + diff_headline_content, 'PH Monitoring:', size=20) # ha='center', va='center', size=20) axbox_desired_value = plt.axes([widgets_x, monitoring_y, textbox_length0, button_height]) axbox_range_warning = plt.axes([widgets_x2, monitoring_y, textbox_length0, button_height]) ## Sampling and plotting text = self.fig.text(headlines_x, output_y + diff_headline_content, 'Output Generation:', size=20) # ha='center', va='center', size=20) axbox_sample_period = plt.axes([widgets_x, chart_s_y, textbox_length1, button_height]) axbox_plot_period = plt.axes([widgets_x2, chart_s_y, textbox_length1, button_height]) axbox_Nsamples_show = plt.axes([widgets_x, chart_s_y2, textbox_length1, button_height]) ax_start = plt.axes([widgets_x,chart_start_stop_y, 0.04, button_height]) ax_stop = plt.axes([widgets_x2, chart_start_stop_y, 0.04, button_height]) ## Output text = self.fig.text(headlines_x, chart_s_y + diff_headline_content, 'Sampling and plotting:', size=20) # ha='center', va='center', size=20) axsave_disk = plt.axes([w1_x, output_y, 0.055, button_height]) axsave_DDBB = plt.axes([w2_x, output_y, 0.055, button_height]) axreport = plt.axes([w3_x, output_y, 0.055, button_height]) ################## Add functionalities ########################### ################ Chart AXES ################: bstop = Button(ax_stop, 'Stop') bstop.on_clicked(self.stop_reading_data) bstart = Button(ax_start, 'Start') bstart.on_clicked(self.start_reading_data) # bprev.on_clicked(self.auto_update_test) #### Text input Period #### initial_text = str(int(self.period_sampling * 1000)); text_box_sample_period = TextBox(axbox_sample_period, 'Sample(ms) ', initial=initial_text) text_box_sample_period.on_submit(self.submit_sample_period) initial_text = str(int(self.period_plotting * 1000)); text_box_plotting_period = TextBox(axbox_plot_period, 'Plot(ms) ', initial=initial_text) text_box_plotting_period.on_submit(self.submit_plotting_period) #### Text input N samples #### initial_text = str(int(self.show_window)); text_Nsamples_show = TextBox(axbox_Nsamples_show, 'Samples Chart ', initial=initial_text) text_Nsamples_show.on_submit(self.submit_show_window) ################ Data generation widgets ################ bpsave_disk = Button(axsave_disk, 'Save Disk') bpsave_disk.on_clicked(self.save_to_disk) bpsave_DDBB = Button(axsave_DDBB, 'Save DDBB') bpsave_DDBB.on_clicked(self.send_buffer_to_DDBB) bpsave_report = Button(axreport, 'Report') bpsave_report.on_clicked(self.generate_report) ################ Cleaning input widgets ################ ## Text input MAchine ID initial_text = self.machine_ID text_box_machine = TextBox(axbox_machineID, 'Machine ID ', initial=initial_text) text_box_machine.on_submit(self.submit_machineID) initial_text = self.piping_ID text_box_piping = TextBox(axbox_pipingID, 'Piping ID ', initial=initial_text) text_box_piping.on_submit(self.submit_pipingID) ################ MONITORING variables ################ initial_text = str(self.Monitor.desired_value); text_desired_value = TextBox(axbox_desired_value, 'Desired PH ', initial=initial_text) text_desired_value.on_submit(self.submit_desired_value) initial_text = str(self.Monitor.range_warning); text_range_warning = TextBox(axbox_range_warning, 'Warning Range ', initial=initial_text) text_range_warning.on_submit(self.submit_range_warning) # I think we needed to keep them in memory of they would die self.buttons = [bstart, bstop, bpsave_disk,bpsave_DDBB,text_box_machine, text_box_sample_period,text_box_plotting_period, text_Nsamples_show, text_desired_value, text_range_warning,bpsave_report, text_box_piping] self.initial_text_data = gl.add_text(positionXY = [0.35,0.5], text = r'Waiting for data',fontsize = 30, ax = data_axes) gl.subplots_adjust(left=.09, bottom=.20, right=.90, top=.90, wspace=.20, hspace=0) self.monitoring_y = monitoring_y
df.to_csv('./out.csv', sep=',') ###### SET THE READING ######## ser = serial.Serial('/dev/ttyUSB0', 9600) ser.readline() for i in range(10): print (float(ser.readline().decode("utf-8").split("\n")[0])) #ser.close() ###### GENERATE FAKE DATA ############ data = np.random.randn(100,1) + 35 time = range(data.size) ###### GENERATE THE FIGURE ############ fig = gl.init_figure(); data_axes = gl.subplot2grid((1,4), (0,0), rowspan=1, colspan=3) data = [] time = [] update_data.index = 0 print ("starting...") ## Define the class with all the info class information(): ## Serial port info serial = ser # Serial port we get the info from rt = None ## Data information
verbose = verbose, time_profiling = time_profiling) ### Set the initial parameters theta_init = None model_theta_init = None ############# PERFORM THE EM ############# logl,theta_list,model_theta_list = myEM.fit(Xdata, model_theta_init = model_theta_init, theta_init = theta_init) spf.print_final_clusters(myDManager,clusters_relation, theta_list[-1], model_theta_list[-1]) ####################################################################################################################### #### Plot the evolution of the centroids likelihood ! ##################################################### ####################################################################################################################### gl.init_figure() gl.plot(range(1,np.array(logl).flatten()[1:].size +1),np.array(logl).flatten()[1:], legend = ["EM LogLikelihood"], labels = ["Convergence of LL with generated data","Iterations","LL"], lw = 2) gl.savefig(folder_images +'Likelihood_Evolution. K_G:'+str(K_G)+ ', K_W:' + str(K_W) + ', K_vMF:' + str(K_vMF)+ '.png', dpi = 100, sizeInches = [12, 6]) if(perform_HMM_after_EM): Ninit = 1 ############# Create the EM object and fit the data to it. ############# clusters_relation = "MarkovChain1" # MarkovChain1 independent myEM = CEM.CEM( distribution = myDManager, clusters_relation = clusters_relation, T = T, Ninit = Ninit, delta_ll = delta_ll, verbose = verbose, time_profiling = time_profiling)
def create_Bayesian_analysis_charts_simplified(model, train_dataset, validation_dataset, tr_loss, val_loss, KL_loss, folder_images, epoch_i=None): # Configurations of the plots alpha_points = 0.2 color_points_train = "dark navy blue" color_points_val = "amber" color_train_loss = "cobalt blue" color_val_loss = "blood" color_truth = "k" color_mean = "b" color_most_likey = "y" ################################ Divide in plots ############################## gl.init_figure() ax1 = gl.subplot2grid((6, 3), (0, 0), rowspan=3, colspan=1) ax2 = gl.subplot2grid((6, 3), (3, 0), rowspan=3, colspan=1, sharex=ax1, sharey=ax1) ax3 = gl.subplot2grid((6, 3), (0, 1), rowspan=2, colspan=1) ax4 = gl.subplot2grid((6, 3), (2, 1), rowspan=2, colspan=1, sharex=ax3) ax5 = gl.subplot2grid((6, 3), (4, 1), rowspan=2, colspan=1, sharex=ax3) ax6 = gl.subplot2grid((6, 3), (0, 2), rowspan=3, colspan=1) ax7 = gl.subplot2grid((6, 3), (3, 2), rowspan=3, colspan=1, sharex=ax6) ####### ax1, ax2: Get confusion matrices ########## labels_classes, confusion = model.get_confusion_matrix(train_dataset) plot_confusion_matrix(confusion, labels_classes, ax1) labels_classes, confusion = model.get_confusion_matrix(validation_dataset) plot_confusion_matrix(confusion, labels_classes, ax2) ############## ax3 ax4 ax5: Loss Evolution !! ###################### ## ax3: Evolutoin of the data loss gl.plot([], tr_loss, ax=ax3, lw=3, labels=["Losses", "", "Data loss (MSE)"], legend=["train"], color=color_train_loss) gl.plot([], val_loss, ax=ax3, lw=3, legend=["validation"], color=color_val_loss, AxesStyle="Normal - No xaxis") ## ax4: The evolution of the KL loss gl.plot([], KL_loss, ax=ax4, lw=3, labels=["", "", "KL loss"], legend=["Bayesian Weights"], AxesStyle="Normal - No xaxis", color="k") ## ax5: Evolutoin of the total loss gl.plot([], tr_loss, ax=ax5, lw=3, labels=["", "epoch", "Total Loss (Bayes)"], legend=["train"], color=color_train_loss) gl.plot([], val_loss, ax=ax5, lw=3, legend=["validation"], color=color_val_loss) ############## ax6 ax7: Variational Weights !! ###################### create_plot_variational_weights(model, ax6, ax7) gl.set_zoom(ax=ax6, ylim=[-0.1, 10]) gl.set_zoom(ax=ax7, xlim=[-2.5, 2.5], ylim=[-0.1, 0.5]) # Set final properties and save figure gl.set_fontSizes(ax=[ax1, ax2, ax3, ax4, ax5, ax6, ax7], title=20, xlabel=20, ylabel=20, legend=10, xticks=12, yticks=12) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.10) if (type(epoch_i) == type(None)): gl.savefig(folder_images + 'Training_Example_Data_Bayesian.png', dpi=100, sizeInches=[20, 10]) else: gl.savefig(folder_images + '%i.png' % epoch_i, dpi=100, sizeInches=[20, 10], close=True, bbox_inches="tight")
def visualize_attention_matrix(question_tokens, passage_tokens, attention_matrix, image_path): """ Text to visualze attention map for.a given exmaple. question_tokens: List of tokens of the question passage_tokens: List of tokens of the passage attention_matrix: len(passage) x len(question) matrix with the probabilities """ f = gl.init_figure() ax = f.add_axes([0.1, 0.3, 0.8, 0.5]) ax_attention_words = f.add_axes([0.1, 0.70, 0.8, 0.15]) ax_attention_words.axis('off') # add image cmap = "binary" #cm.get_cmap('coolwarm', 30) i = ax.imshow(attention_matrix, interpolation='nearest', cmap=cmap, vmin=0, vmax=1) # add colorbar cbaxes = f.add_axes([0.95, 0.3, 0.02, 0.5]) cbar = f.colorbar(i, cax=cbaxes, orientation='vertical') cbar.ax.set_xlabel('Probability', labelpad=6) # add labels ax.set_yticks(range(len(question_tokens))) ax.set_yticklabels(question_tokens) ax.set_xticks(range(len(passage_tokens))) ax.set_xticklabels(passage_tokens, rotation=80) ax.set_xlabel('Passage') ax.set_ylabel('Question') ########### GET THE MOST ATTENTION WORDS ######## Nmax_attention_words = 3 z = (-attention_matrix).argsort(axis=1)[:, :] attentioned_passage_words = [] for i in range(len(question_tokens)): attentioned_passage_words.append([]) for j in range(Nmax_attention_words): attentioned_passage_words[-1].append( passage_tokens[z[i, j]] + "(%.1f%%)" % (attention_matrix[i, z[i, j]] * 100)) attentioned_passage_words[-1] = ", ".join( attentioned_passage_words[-1]) text_correspondance = "" for i in range(len(question_tokens)): text_correspondance += question_tokens[ i] + " ---> " + attentioned_passage_words[i] + "\n" ax_attention_words.text(0, 0, text_correspondance) # ax2.yaxis.tick_right() # ax2.yaxis.set_label_position("right") f.show() # gl.set_fontSizes(ax = [ax1,ax2], title = 20, xlabel = 15, ylabel = 18, # legend = 12, xticks = 14, yticks = 14) gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.20, hspace=0.10) gl.savefig(image_path, dpi=100, sizeInches=[10, 6], close=False, bbox_inches="tight")
## Comparison threshold_corte_ultimo_escano_list = np.linspace(Total_votos,0,10000) blue_party = total_parties_blue[2] red_party = total_parties_red[1] party_escanhos_blue_2, party_escanhos_red_2 = get_esanhos_between_two_formations(blue_party,red_party, threshold_corte_ultimo_escano_list,N_escanos) """ PLOT THE INDIVIDUAL ESCANHOS AND THE TOTAL IN THE END """ gl.init_figure(); ax1 = gl.subplot2grid((3,1), (0,0), rowspan=1, colspan=1) ax2 = gl.subplot2grid((3,1), (1,0), rowspan=1, colspan=1, sharex = ax1, sharey = ax1) ax3 = gl.subplot2grid((3,1), (2,0), rowspan=1, colspan=1, sharex = ax1, sharey = ax1) gl.plot(threshold_corte_ultimo_escano_list_blue,party_escanhos_blue, ax = ax1, labels = ["Numero de escanhos obtenidos en funcion del numero de votos del ultimo escanho", "Numero de votos del ultimo escanho", "Numero de escanhos obtenidos"], legend = [["%.2f"%(total_parties_blue[i][j]) for j in range(len(total_parties_blue[i]))] for i in range(len(total_parties_blue))]) gl.plot(threshold_corte_ultimo_escano_list_red,party_escanhos_red, ax = ax2, labels = ["Numero de escanhos obtenidos en funcion del numero de votos del ultimo escanho", "Numero de votos del ultimo escanho", "Numero de escanhos obtenidos"], legend = [["%.2f"%(total_parties_red[i][j]) for j in range(len(total_parties_red[i]))] for i in range(len(total_parties_red))]) gl.plot(threshold_corte_ultimo_escano_list,party_escanhos_blue_2, ax = ax3, labels = ["", "", ""],
###### SET THE READING ######## if (0): ser = serial.Serial('/dev/ttyUSB0', 9600) ser.readline() for i in range(10): print(float(ser.readline().decode("utf-8").split("\n")[0])) #ser.close() ###### GENERATE FAKE DATA ############ data = np.random.randn(100, 1) + 35 time = range(data.size) ###### GENERATE THE FIGURE ############ fig = gl.init_figure() data_axes = gl.subplot2grid((1, 4), (0, 0), rowspan=1, colspan=3) data = [] time = [] update_data.index = 0 print("starting...") ## Define the class with all the info class information(): ## Serial port info serial = ser # Serial port we get the info from rt = None
classifiers_keys = cl_d.keys() Nclassifiers = len(classifiers_keys) for key in classifiers_keys: classifier = cl_d[key] train_acc.append(get_class_rate(Ytrain, classifier.predict(Xtrain))) test_acc.append(get_class_rate(Ytest, classifier.predict(Xtest))) train_CE.append(get_CE(Ytrain, classifier.predict_proba(Xtrain)[:, 1])) test_CE.append(get_CE(Ytest, classifier.predict_proba(Xtest)[:, 1])) train_acc = np.array(train_acc) test_acc = np.array(test_acc) train_CE = np.array(train_CE) test_CE = np.array(test_CE) gl.init_figure() ax1 = plt.subplot(2, 1, 1) plt.bar(np.arange(Nclassifiers) + 0.2, 1 - train_acc, width=0.2, color='c', align='center', label="train") plt.bar(np.arange(Nclassifiers) + 0.4, 1 - test_acc, width=0.2, color='r', align='center', label="test") plt.xticks(np.arange(Nclassifiers) + 0.3, classifiers_keys) plt.title('Classifiers Performance')
def create_image_weights_epoch(model, video_fotograms_folder2, epoch_i): """ Creates the image of the training and validation accuracy """ N_Bayesian_layers = len(model.VBmodels) N_Normal_layers = len(model.LinearModels) # Compute the number of squares we will need: # 1 x linear layers, 2 x LSTMS gl.init_figure(); cmap = cm.get_cmap('coolwarm', 30) all_axes = [] for i in range(N_Bayesian_layers): layer = model.VBmodels[i] # if (layer.type_layer == "linear"): if ("linear" in type(layer).__name__.lower()): ax = gl.subplot2grid((1,N_Bayesian_layers + N_Normal_layers), (0,i), rowspan=1, colspan=1) weights = layer.weight.detach().cpu().numpy() biases = layer.bias.detach().cpu().numpy().reshape(-1,1) neurons = np.concatenate((weights, biases), axis = 1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) else: ax = gl.subplot2grid((1,N_Bayesian_layers + N_Normal_layers), (0,i), rowspan=1, colspan=1) weights_ih = layer.weight_ih.detach().cpu().numpy() biases_ih = layer.bias_ih.detach().cpu().numpy().reshape(-1,1) weights_hh = layer.weight_hh.detach().cpu().numpy() biases_hh = layer.bias_hh.detach().cpu().numpy().reshape(-1,1) weights = np.concatenate((weights_ih,weights_hh),axis = 1) biases = np.concatenate((biases_ih,biases_hh),axis = 1) neurons = np.concatenate((weights, biases), axis = 1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) for i in range(N_Normal_layers): layer = model.LinearModels[i] if ("linear" in type(layer).__name__.lower()): ax = gl.subplot2grid((1,N_Bayesian_layers + N_Normal_layers), (0,N_Bayesian_layers +i), rowspan=1, colspan=1) weights = layer.weight.detach().cpu().numpy() biases = layer.bias.detach().cpu().numpy().reshape(-1,1) neurons = np.concatenate((weights, biases), axis = 1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) else: ax = gl.subplot2grid((1,N_Bayesian_layers + N_Normal_layers), (0,N_Bayesian_layers +i), rowspan=1, colspan=1) weights_ih = layer.weight_ih.detach().cpu().numpy() biases_ih = layer.bias_ih.detach().cpu().numpy().reshape(-1,1) weights_hh = layer.weight_hh.detach().cpu().numpy() biases_hh = layer.bias_hh.detach().cpu().numpy().reshape(-1,1) weights = np.concatenate((weights_ih,weights_hh),axis = 1) biases = np.concatenate((biases_ih,biases_hh),axis = 1) neurons = np.concatenate((weights, biases), axis = 1) cax = ax.imshow(neurons, interpolation="nearest", cmap=cmap, vmin=-2, vmax=2) all_axes.append(ax) # plt.xticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='vertical') # plt.yticks(range(data_df_train.shape[1]), data_df_train.columns, rotation='horizontal') plt.colorbar(cax) # plt.colorbar(cax2) # ax1.set_xticks(data_df_train.columns) # , rotation='vertical' # ax1.grid(True) plt.title('Weights ') # labels=[str(x) for x in range(Nshow )] # ax1.set_xticklabels(labels,fontsize=20) # ax1.set_yticklabels(labels,fontsize=20) # Add colorbar, make sure to specify tick locations to match desired ticklabels plt.show() gl.set_fontSizes(ax = [all_axes], title = 20, xlabel = 20, ylabel = 20, legend = 20, xticks = 12, yticks = 12) # Set final properties and save figure gl.subplots_adjust(left=.09, bottom=.10, right=.90, top=.95, wspace=.30, hspace=0.30) gl.savefig(video_fotograms_folder2 +'%i.png'%epoch_i, dpi = 100, sizeInches = [14, 10], close = True, bbox_inches = None)