import numpy as np from bokeh.plotting import figure, output_file, save # , show from tensorboard.backend.event_processing import event_accumulator file_path = "/tigress/alexeys/worked_Graphs/Graph16_momSGD_new/" ea1 = event_accumulator.EventAccumulator( file_path + "events.out.tfevents.1502649990.tiger-i19g10") ea1.Reload() ea2 = event_accumulator.EventAccumulator( file_path + "events.out.tfevents.1502652797.tiger-i19g10") ea2.Reload() histograms = ea1.Tags()['histograms'] # ages': [], 'audio': [], 'histograms': ['input_2_out', # 'time_distributed_1_out', 'lstm_1/kernel_0', 'lstm_1/kernel_0_grad', # 'lstm_1/recurrent_kernel_0', 'lstm_1/recurrent_kernel_0_grad', # 'lstm_1/bias_0', 'lstm_1/bias_0_grad', 'lstm_1_out', 'dropout_1_out', # 'lstm_2/kernel_0', 'lstm_2/kernel_0_grad', 'lstm_2/recurrent_kernel_0', # 'lstm_2/recurrent_kernel_0_grad', 'lstm_2/bias_0', 'lstm_2/bias_0_grad', # 'lstm_2_out', 'dropout_2_out', 'time_distributed_2/kernel_0', # 'time_distributed_2/kernel_0_grad', 'time_distributed_2/bias_0', # 'time_distributed_2/bias_0_grad', 'time_distributed_2_out'], 'scalars': # ['val_roc', 'val_loss', 'train_loss'], 'distributions': ['input_2_out', # 'time_distributed_1_out', 'lstm_1/kernel_0', 'lstm_1/kernel_0_grad', # 'lstm_1/recurrent_kernel_0', 'lstm_1/recurrent_kernel_0_grad', # 'lstm_1/bias_0', 'lstm_1/bias_0_grad', 'lstm_1_out', 'dropout_1_out', # 'lstm_2/kernel_0', 'lstm_2/kernel_0_grad', 'lstm_2/recurrent_kernel_0', # 'lstm_2/recurrent_kernel_0_grad', 'lstm_2/bias_0', 'lstm_2/bias_0_grad', # 'lstm_2_out', 'dropout_2_out', 'time_distributed_2/kernel_0', # 'time_distributed_2/kernel_0_grad', 'time_distributed_2/bias_0', # 'time_distributed_2/bias_0_grad', 'time_distributed_2_out'], 'tensors': # [], 'graph': True, 'meta_graph': True, 'run_metadata': []} for h in histograms: x1 = np.array(ea1.Histograms(h)[0].histogram_value.bucket_limit[:-1]) y1 = ea1.Histograms(h)[0].histogram_value.bucket[:-1] x2 = np.array(ea2.Histograms(h)[0].histogram_value.bucket_limit[:-1]) y2 = ea2.Histograms(h)[0].histogram_value.bucket[:-1] h = h.replace("/", "_") p = figure(title=h, y_axis_label="Arbitrary units", x_axis_label="Arbitrary units") # , y_axis_type="log") p.line(x1, y1, legend="float16, SGD with momentum", line_color="green", line_width=2) p.line(x2, y2, legend="float32, SGD with momentum", line_color="indigo", line_width=2) p.legend.location = "top_right" output_file("plot" + h + ".html", title=h) save(p) # open a browser