小説を読もうの累計ランキングをDoc2Vecで解析する その４

import numpy as np
import matplotlib.pyplot as plt
from gensim.models.doc2vec import Doc2Vec
from sklearn.manifold import TSNE

# 学習済みモデルを読み込む
m = Doc2Vec.load('drive/My Drive/Colab Notebooks/syosetu/doc2vec.model')
weights = []
for i in range(0, len(m.docvecs)):
  weights.append(m.docvecs[i].tolist())
weights_tuple = tuple(weights)
X = np.vstack(weights_tuple)

# t-SNEで次元圧縮する
tsne_model = TSNE(n_components=2, random_state=0, verbose=2)
np.set_printoptions(suppress=True)
t_sne = tsne_model.fit_transform(X)

# クラスタリング済みのデータを読み込む
with open('drive/My Drive/Colab Notebooks/syosetu/novel_cluster.csv', 'r') as f:
  reader = csv.reader(f)
  clustered = np.array([row for row in reader])
  clustered = clustered.astype(np.dtype(int).type)
  clustered = clustered[np.argsort(clustered[:, 0])]
  clustered = clustered.T[1]

# グラフ描画
fig, ax = plt.subplots(figsize=(10, 10), facecolor='w', edgecolor='k')

# Set Color map
cmap = plt.get_cmap('Dark2')

for i in range(t_sne.shape[0]):
  cval = cmap(clustered[i] / 4)
  ax.scatter(t_sne[i][0], t_sne[i][1], marker='.', color=cval)
  ax.annotate(i, xy=(t_sne[i][0], t_sne[i][1]), color=cval)
plt.show()

!pip install gensim torch tensorboardX

from torch import FloatTensor
from gensim.models import KeyedVectors
from tensorboardX import SummaryWriter

m = KeyedVectors.load('drive/My Drive/Colab Notebooks/syosetu/doc2vec.model')
weights = []
labels = []
for i in range(0, len(m.docvecs)):
  weights.append(m.docvecs[i].tolist())
  labels.append(m.docvecs.index_to_doctag(i))

# DEBUG: visualize vectors up to 1000
weights = weights[:1000]
labels = labels[:1000]

writer = SummaryWriter()
writer.add_embedding(FloatTensor(weights), metadata=labels)