graphing.py 文件源码

def sentence_length_vs_ponder_time(config,processed_data):

    plotting_data = []

    for  data in processed_data:

        steps = len(data["act_probs"])
        hyp_length = sum([1 for x in data["hypothesis"] if x!="PAD"])
        prem_length = sum([1 for x in data["premise"] if x!="PAD"])
        avg_length = (hyp_length + prem_length)/2
        correct = data["correct"]
        type_class = data["class"]
        plotting_data.append([steps, avg_length, correct, type_class])


    plotting_data = pd.DataFrame(np.vstack(plotting_data), columns=["steps", "avg_length", "correct", "class"])

    seaborn.violinplot(x="steps", y="avg_length",
                       hue="correct",split=True,
                       data=plotting_data, inner="quartile", scale="count")
    plt.show()
    # for class_type in [0.0,1.0,2.0]:
    #     fig = plt.figure()
    #     x_vals = [x[0] for x in plotting_data if (x[3]==class_type and x[2]==1.0)]
    #     y_vals = [x[1] for x in plotting_data if (x[3]==class_type and x[2]==1.0)]
    #     print("Class: ",class_type, "No. Correct: ", len(x_vals))
    #     plt.scatter(x_vals, y_vals,color="g")
    #
    #     x_vals = [x[0] for x in plotting_data if (x[3]==class_type and x[2]==0.0)]
    #     y_vals = [x[1] for x in plotting_data if (x[3]==class_type and x[2]==0.0)]
    #     print("Class: ",class_type, "No. Incorrect: ", len(x_vals))
    #
    #     plt.scatter(x_vals, y_vals,color="r")
    #
    #     ax = plt.gca()
    #     ax.set_xlabel("ACT Steps")
    #     ax.set_ylabel("avg hyp/premise length")
    #     ax.set_title("test_title")
    #     plt.show()