seaborn/matplotlib 中的直方圖顯示 x 軸上的所有分箱數據索引

Question

我需要制作給定列表中所有值的直方圖。 我使用了 seaborn 的 distplot，但其中一個軸沒有顯示從 0 到列表中最后一個元素的索引，而是顯示了某種形式的分布。

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

sns.set(style="whitegrid")
print(sns.__version__)

sns.set_context(font_scale=1.5)

data = [1.1992198, 1.2429917, 0.7515156, 1.7279389, -0.16715668, 0.94465995, 0.5149495, 0.94465995, 0.94465995, 3.8740897, 7.453125, 7.453125, 1.0201894, 1.444468, 0.06495813, 0.18581325, -0.69003785, 3.1213043, 0.24899049, -0.5395518, 2.6421795, 2.238052, -0.42627642, 0.689369, 1.0177083, 0.0021173293, 0.19708821, -0.6978323, -0.27355388, -1.0527502, -1.2287112, -0.73426425, -1.5779951, -1.4275085, -0.72636086, 0.49798694, 0.5233074, -0.8736689, -1.5343369, 0.83868057, 0.14993721, -1.5746347, -1.1844425, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328, 1.9799328]

data2 = [0.34094468, -1.8498722, -0.35345584, -1.0018779, 0.18884292, -2.6028345, 0.39934048, -0.119069986, -0.20210052, -0.2972668, -0.8640028, -0.6174464, 0.096682094, -1.9147822, -0.7738649, 0.6141649, 0.86409974, -0.5216787, -0.78182876, 0.22742827, -0.840597, -0.97359276, -0.018100848, -0.5059276, -1.7152423, -0.07815174, 0.18345535, -0.76344514, -0.39645284, 0.18889628, -0.5543669, -0.18788649, -0.13553666, 0.1379985, 0.65224963, 0.5777133, -0.9204392, -0.91472155, -0.58848035, -1.6883624, -0.58383256, 0.25340325, -0.09143271, 0.50240713, 0.8944117, 0.07218201, 1.1128205, 1.3817745, -0.09530114, 0.56783175, -0.12059356, 0.43868077, -0.2728266, 0.61756617, -0.51779836, -0.39096248, -0.635239, -0.635239, -0.5384383, -0.635239, -0.6920986, -0.9351034, -0.9254051, -0.842712, -1.1218141]

data3 = [0.72135484, -0.706092, 0.36165744, 0.40211153, 0.14495818, 0.9395333, 1.450367, 0.32213485, 0.52471924, 3.3083296, 6.7051606, 6.1889296, -0.210258, -0.09990394, -0.85894525, -0.36614275, -1.5075212, 1.8715478, 0.29819223, -1.0022302, 2.108101, 1.8913394, -0.24430388, 0.059003413, -0.39443398, -0.0057572527, 0.5327027, -1.4999104, -0.60988855, -0.95330614, -1.9033353, -0.93208313, -1.7135317, -1.2041125, 0.007865965, 1.0655571, -0.42969102, -1.9678588, -2.165072, -1.1763439, -0.4736237, -0.8522189, -1.073197, 2.3406122, 2.8758054, 1.9956598, 3.3263054, 3.0907226, 1.8059512, 2.533312, 2.1851382, 2.1604633, 1.7256155, 2.912341, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354, 2.0519354]

sns.set_style("whitegrid")

# bins don't seem to do anything here
sns.distplot(data, norm_hist=0, hist=False, kde_kws={"shade":True, "bw": 0.05}, bins=100, color="b", label="attn")
sns.distplot(data2, norm_hist=0, hist=False, kde_kws={"shade":True, "bw": 0.05}, bins=100, color="g", label="attn_rel_pos")
pls = sns.distplot(data3, norm_hist=0, hist=False, kde_kws={"shade":True, "bw": 0.05}, bins=100, color="r", label="attn_comb")
plt.legend()


pls.axes.set_title("Title")
pls.set(xlabel='Show indexes from 0 to last here', ylabel='My Weight')

結果如下：

我需要展示列表中的權重如何從列表的 0 索引到最后一個（所有三個列表的長度相同）分布。 由於列表的長度約為 60 個元素，因此我可能會將它們裝箱，但我也找不到任何在這里真正起作用的 bin 參數。

這是我需要的photoshopped版本（但我還需要正確呈現數據，而不是某種形式的密集分布）：

在這一點上，我也不關心它是否是 seaborn，如果在 matplotlib 中更容易做到，那么我也可以接受該解決方案。 非常感謝！

編輯：值總是高於 0 並使用條形圖而不是直方圖的示例，是否可以將值顯示為連續線：

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

sns.set(style="whitegrid")
# print(sns.__version__)

fig = plt.figure(figsize=(8, 6))
# Not showing the data lists here

def softmax(x):
    return np.exp(x) / np.exp(x).sum(axis=0)

# attn
data = [1.2227119, 2.3106432, 0.3175479, 2.6548655, 0.87468535, -2.5941365, -2.8492305, -2.5941365, -2.5941365, 0.78577393, -3.1803715, -3.1803715, -2.527419, -0.32336473, -0.23149578, 0.1958992, -1.1464257, -1.3171446, -0.82610035, -0.6265811, -0.04922826, 1.268781, -0.63436747, 1.6067829, -0.12655944, 0.30039954, -0.16766489, -2.2401857, -0.036131226, -0.22972624, -0.041365635, 0.6901127, -1.3901691, -0.87032473, 0.13755159, 0.013177752, 1.343483, 0.17142272, -0.08306693, 0.9223409, -0.43641013, -1.2699138, -1.0307136, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937, -0.91284937]

# attn_pos
data2 = [ 0.41763785, -2.3683536, -0.10784631, -0.18569931, 0.20513327, -2.017535, -0.13072428, -1.423852, -0.037444096, -0.29240212, -0.33224604, -0.8710863, -0.2023627, -0.72792727, -0.15152885, 0.20426698, 0.9366691, -0.04617013, 0.03217609, -1.2339046, -0.54832625, -1.0488682, -0.13601255, 0.46882343, -1.1635672, -0.98546046, -0.5023038, -1.7461116, -0.38710907, -0.75636834, -0.077942476, -1.5930176, -0.8428734, 0.36083415, -0.235654, 0.27319083, 0.2481483, -0.8994429, 0.51432747, 0.3907901, -1.0105598, -0.3258296, -0.53897303, 1.3885188, -0.8589748, -0.25751373, 0.9007931, 0.64612645, -0.9111746, -0.5629407, 0.5113405, -0.5913975, 0.8435228, -0.0041873083, 1.2919891, 0.49669626, 0.96742135, 0.96742135, 0.9375946, 0.96742135, 0.62772936, 0.2486542, 0.59175867, 0.13839966, 0.33540285]

# combined
data3 = [1.2788991, 1.8618634, -0.39062586, 2.4721837, 3.1994174, -2.539834, -1.7466183, -2.7143257, -2.5147946, 2.0141108, -3.282056, -3.381072, -3.5476263, -0.7556571, -0.9092123, 0.39682359, -0.7735753, -0.5908251, -1.0364372, -1.0283178, -0.7909626, 1.2081728, -0.96530867, 2.3020573, 0.84674144, 0.7156407, 0.1626791, -1.6639395, 0.27049372, 0.5723161, -0.89840394, 1.5462611, -2.1371794, -1.70861, 0.30733263, 0.22821441, 0.7041679, -0.36799663, 0.27130017, 0.98303056, -1.142178, -2.035885, -1.7442997, -0.49790236, -1.6333843, -0.09258777, -1.971946, -1.0179313, -1.1023216, -0.7061392, -1.0784137, -1.8746437, -0.015198052, -0.9757373, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813, -0.43436813]

softmax_flag = True

if softmax_flag:
    plt.bar(range(len(data)), softmax(data), color='b', alpha=0.3, label ='attn') #, hatch="/")
    plt.bar(range(len(data2)), softmax(data2), color='g', alpha = 0.3, label ='attn_rel_pos') #, hatch="o")
    plt.bar(range(len(data3)), softmax(data3), color='r',alpha = 0.3, label ='attn_comb') #, hatch="\\")
    plt.plot(range(len(data)), softmax(data),'bx', alpha = 0.5)
    plt.plot(range(len(data2)), softmax(data2),'go', alpha = 0.5, ms = 4)
    plt.plot(range(len(data3)), softmax(data3),'r+', alpha = 0.5)
else:
    plt.bar(range(len(data)), data, color='b', alpha=0.3, label ='attn') #, hatch="/")
    plt.bar(range(len(data2)), data2, color='g', alpha = 0.3, label ='attn_rel_pos') #, hatch="o")
    plt.bar(range(len(data3)), data3, color='r',alpha = 0.3, label ='attn_comb') #, hatch="\\")
    plt.plot(range(len(data)), data,'bx', alpha = 0.5)
    plt.plot(range(len(data2)), data2,'go', alpha = 0.5, ms = 4)
    plt.plot(range(len(data3)), data3,'r+', alpha = 0.5)

plt.title("Head 3")
plt.xlabel("Word Position in the Sentence")
plt.ylabel("Attention Weight")

# plt.grid()
plt.legend(fontsize=16)

plt.savefig('head_3_in_32_softmax.png', dpi=350)

# plt.close(fig)

圖片：

Answer 1

我不確定您是否希望在每個索引值處都有某種離散峰值作為一種直方圖。 但從你的問題我想以下是你想要的：

plt.bar(range(len(data)), data, color='b', alpha=0.3, label ='attn')
plt.bar(range(len(data2)), data2, color='g', alpha = 0.3, label ='attn_rel_pos')
plt.bar(range(len(data3)), data3, color='r',alpha = 0.3, label ='attn_comb')

輸出

如果您想以某種方式區分同一索引的每個高度/值，除了條形圖之外，您還可以使用一些標記，如下所示（這里我沒有使用sns ）：

import matplotlib.pyplot as plt

fig = plt.figure(figsize=(8, 6))
# Not showing the data lists here

plt.bar(range(len(data)), data, color='b', alpha=0.3, label ='attn') #, hatch="/")
plt.bar(range(len(data2)), data2, color='g', alpha = 0.3, label ='attn_rel_pos') #, hatch="o")
plt.bar(range(len(data3)), data3, color='r',alpha = 0.3, label ='attn_comb') #, hatch="\\")
plt.plot(range(len(data)), data,'bx', alpha = 0.5)
plt.plot(range(len(data2)), data2,'go', alpha = 0.5, ms = 4)
plt.plot(range(len(data3)), data3,'r+', alpha = 0.5)
plt.grid()
plt.legend(fontsize=16)

輸出

如果您希望它們作為連續線，您可以使用以下內容：

import numpy as np
import matplotlib.pyplot as plt

fig = plt.figure(figsize=(8, 6))
ax = fig.add_subplot(111)

def softmax(x):
    return np.exp(x) / np.exp(x).sum(axis=0)

# Not showing the data lists here

softmax_flag = True

if softmax_flag:
    plt.plot(range(len(data)), softmax(data),'-b', alpha = 0.5, label ='attn')
    plt.plot(range(len(data2)), softmax(data2),'-g', alpha = 0.5, lw=2, label ='attn_rel_pos')
    plt.plot(range(len(data3)), softmax(data3),'-r', alpha = 0.5, lw=2, label ='attn_comb')
    ax.fill_between(range(len(data)), 0, softmax(data), color='dodgerblue', alpha = 0.4)
    ax.fill_between(range(len(data)), 0, softmax(data2), color='mediumseagreen', alpha = 0.4)
    ax.fill_between(range(len(data)), 0, softmax(data3), color='indianred', alpha = 0.4)   
else:
    plt.bar(range(len(data)), data, color='b', alpha=0.3, label ='attn') #, hatch="/")
    plt.bar(range(len(data2)), data2, color='g', alpha = 0.3, label ='attn_rel_pos') #, hatch="o")
    plt.bar(range(len(data3)), data3, color='r',alpha = 0.3, label ='attn_comb') #, hatch="\\")
    plt.plot(range(len(data)), data,'bx', alpha = 0.5)
    plt.plot(range(len(data2)), data2,'go', alpha = 0.5, ms = 4)
    plt.plot(range(len(data3)), data3,'r+', alpha = 0.5)

plt.title("Head 3")
plt.xlabel("Word Position in the Sentence")
plt.ylabel("Attention Weight")

plt.grid()
plt.legend(fontsize=16)

輸出