def compare_per_nt(file1, file2, y_label='New'):
nt1, data1 = list(read_jsons(file1))
nt2, data2 = list(read_jsons(file2))
assert nt1 == nt2
nt = nt1
data1 = np.array(data1)
data2 = np.array(data2)
diff = data2 - data1
ind = np.arange(len(nt))
p1 = plt.bar(ind, data1)
for bar in p1:
bar.set_facecolor(COLOR_BASE)
p2 = plt.bar(ind, diff, bottom=data1)
for id, bar in enumerate(p2):
if diff[id] >= 0:
bar.set_facecolor(COLOR_GREEN)
else:
bar.set_facecolor(COLOR_RED)
custom_lines = [Line2D([0], [0], color=COLOR_BASE, lw=4),
Line2D([0], [0], color=COLOR_GREEN, lw=4),
Line2D([0], [0], color=COLOR_RED, lw=4)]
plt.legend(custom_lines, ['База', 'Улучшение', 'Ухудшение'])
# plt.legend((p1[0], p2[0]), (file1, file2))
add_nt_x_ticks(nt)
plt.ylabel(y_label, fontsize=12)
plt.show()
def compare_per_nt_diff_only(file1, file2, y_label='New'):
nt1, data1 = list(read_jsons(file1))
nt2, data2 = list(read_jsons(file2))
assert nt1 == nt2
nt = read_json('data/ast/non_terminals_plot_modified_attention.json')
assert len(nt1) == len(nt)
data1 = np.array(data1)
data2 = np.array(data2)
diff = data2 - data1
ind = np.arange(len(nt))
p1 = plt.bar(ind, (diff) * 100, width=1)
for id, bar in enumerate(p1):
if diff[id] >= 0:
bar.set_facecolor(COLOR_GREEN)
else:
bar.set_facecolor(COLOR_RED)
custom_lines = [
Line2D([0], [0], color=COLOR_GREEN, lw=4),
Line2D([0], [0], color=COLOR_RED, lw=4)
]
plt.legend(custom_lines, ['Улучшение', 'Ухудшение'], prop={'size': 16})
plt.grid(True)
add_nt_x_ticks(nt)
plt.ylabel(y_label, fontsize=14)
plt.show()
def compare_per_nt(file1, file2, y_label):
nt1, res1 = list(read_jsons(file1))
nt2, res2 = list(read_jsons(file2))
assert nt1 == nt2
x = np.arange(len(nt1))
y1 = np.array(res1)
y2 = np.array(res2)
my_xticks = nt1
plt.xticks(x,
my_xticks,
rotation=30,
horizontalalignment='right',
fontsize=5)
plt.ylabel(y_label)
plt.grid(True)
plt.plot(x, (y2 - y1) * 100)
plt.show()
def print_tree_heights_stats_from_file(tree_heights_file):
print_tree_heights_stats(list(read_jsons(tree_heights_file))[0])
def plot_percentile_from_file(file, x_label, y_label):
stat = list(read_jsons(file))[0]
x, y = get_percentile_plot(stat)
draw_plot(x, y, x_label=x_label, y_label=y_label)
def nearest(x, y, vocab, word):
w_id = -1
for i in range(len(vocab)):
if vocab[i] == word:
w_id = i
if w_id == -1:
raise Exception('No such word in vocabulary: {}'.format(word))
px = x[w_id]
py = y[w_id]
p = np.array([px, py])
points_with_distance = []
for i in range(len(x)):
points_with_distance.append(
(i, np.linalg.norm(np.array([x[i], y[i]]) - p)))
print('Nearest to {}:'.format(vocab[w_id]))
for c_p in sorted(points_with_distance, key=lambda x: x[1])[:10]:
print(vocab[c_p[0]])
if __name__ == '__main__':
emb = Embeddings(
vector_file='/Users/zerogerc/Documents/diploma/GloVe/vectors.txt',
embeddings_size=5)
vocab = list(read_jsons('data/ast/non_terminals.json'))[0]
vocab.append('EOF')
tsne_plot(emb, vocab)
def draw_per_nt_plot(file, y_label='Per NT accuracy'):
nt, data = list(read_jsons(file))
draw_per_nt_plot_inner(nt, Plot(data=data), y_label=y_label)