util.py

import matplotlib.pyplot as plt
from itertools import accumulate
from collections import defaultdict
import numpy as np

def write_list(aer, path):
	f = open(path, 'w')
	for rate in aer:
		f.write(str(rate) + '\n')
	f.close()

def read_list(path):
	f = open(path, 'r')
	return [float(line.split('\n')[0]) for line in f]

def plot_aer(aers, path):
	plt.plot(range(len(aers)), aers)
	plt.savefig(path + 'AER-plot.pdf')
	plt.clf()


def plot_jump(jump, max_jump, path):
	xs = list(map(lambda x : x - max_jump, list(range(len(jump[0])))))
	plt.plot(xs, jump[0])
	plt.savefig(path + 'jump-plot.pdf')
	plt.clf()

def draw_alignment_null(naacl_path, french_path, english_path, fig_path, sure=False, sentence=1):
	"""

	"""
	i = 1
	f = open(french_path, 'r')
	while i <= sentence:
		french = f.readline().split()
		i += 1
	f.close()
	
	i = 1
	e = open(english_path, 'r')
	while i <= sentence:
		english = e.readline().split()
		i += 1
	english = ['NULL']+english
	e.close()

	n = open(naacl_path, 'r')
	alignments = defaultdict(list)
	line = n.readline()
	while line[0] != str(sentence):
		line = n.readline().split()
	while line[0] != str(sentence+1):
		if sure:
			if line[-1] != 'P':
				alignments[int(line[2])-1].append(int(line[1]))
		else:
			alignments[int(line[2])-1].append(int(line[1]))
		line = n.readline().split()
	
	bitext = ([(french, english)], alignments)

	fig = plt.figure(figsize=(8, 4))
	ax = plt.axes()
	plt.axis('off')
	coordinates = get_coordinates(bitext)

	
	lines = [ax.plot(xy[0], xy[1],alpha=0.9,linestyle='-',color='#1a75ff',lw=1.0, solid_capstyle='round')[0] for xy in coordinates['edges']]

	ax.scatter(coordinates['x_f']+coordinates['x_e'], coordinates['y_f']+coordinates['y_e'],
					s=30, c='white', marker='o', lw=0, alpha=1)

	plot_words(ax, coordinates['x_f'], coordinates['y_f'], coordinates['w_f'], 'top')
	plot_words(ax, coordinates['x_e'], coordinates['y_e'], coordinates['w_e'], 'bottom')
	
	plt.savefig(fig_path)

def draw_all_alignments(french_path, english_path, fig_path, sure=False, sentence=1):
	"""
	Draws an alignment that is weighted according to probs of alignment.
	"""
	i = 1
	f = open(french_path, 'r')
	while i <= sentence:
		french = f.readline().split()
		i += 1
	f.close()
	
	i = 1
	e = open(english_path, 'r')
	while i <= sentence:
		english = e.readline().split()
		i += 1
	english = ['NULL']+english
	e.close()

	alignments = defaultdict(list)
	
	bitext = ([(french, english)], alignments)

	fig = plt.figure(figsize=(8, 4))
	ax = plt.axes()
	plt.axis('off')
	coordinates = get_coordinates(bitext, draw_all=True)

	raw_line_weights = []
	
	for j, f in enumerate(french):
	# for i, e in enumerate(english):
		for i, e in enumerate(english):
		# for j, f in enumerate(french):
			raw_line_weights.append(1./len(english))

	# print(len(raw_line_weights))
	line_weights = [w*10 for w in raw_line_weights]
	print(line_weights)
	edge_coords = coordinates['edges']
	print(len(edge_coords))
	lines = [ax.plot(xy[0], xy[1], alpha=0.9, linewidth=w, linestyle='-', color='#1a75ff', solid_capstyle='round')[0] for xy,w in zip(coordinates['edges'], line_weights)]


	ax.scatter(coordinates['x_f']+coordinates['x_e'], coordinates['y_f']+coordinates['y_e'],
					s=30, c='white', marker='o', lw=0, alpha=1)

	plot_words(ax, coordinates['x_f'], coordinates['y_f'], coordinates['w_f'], 'top')
	plot_words(ax, coordinates['x_e'], coordinates['y_e'], coordinates['w_e'], 'bottom')
	
	plt.savefig(fig_path + '.pdf')
	plt.clf()

def draw_weighted_alignment_null(model, naacl_path, french_path, english_path, fig_path, sure=False, sentence=1, all_ones=False):
	"""
	Draws an alignment that is weighted according to probs of alignment.
	"""
	i = 1
	f = open(french_path, 'r')
	while i <= sentence:
		french = f.readline().split()
		i += 1
	f.close()
	
	i = 1
	e = open(english_path, 'r')
	while i <= sentence:
		english = e.readline().split()
		i += 1
	english = ['NULL']+english
	e.close()

	alignments = defaultdict(list)

	bitext = ([(french, english)], alignments)

	fig = plt.figure(figsize=(8, 4))
	ax = plt.axes()
	plt.axis('off')
	coordinates = get_coordinates(bitext, draw_all=True)

	# get weights of the edges:

	F_indices = []
	for i in french:
		try:
			f = model.V_f_indices[i]
		except KeyError:
			f = model.V_f_indices['-UNK-']
		F_indices.append(f)

	E_indices = []
	for i in english:
		try:
			e = model.V_e_indices[i]
		except KeyError:
			e = model.V_e_indices['-UNK-']
		E_indices.append(e)

	raw_line_weights = []
	
	for j, f in enumerate(F_indices):
	# for i, e in enumerate(E_indices):
		posterior = model.posterior(f, j, E_indices, F_indices)
		for i, e in enumerate(E_indices):
		# for j, f in enumerate(F_indices):
			
			p = posterior[i]
			raw_line_weights.append(p)

	print(len(raw_line_weights))
	line_weights = [w*10 for w in raw_line_weights]
	print(line_weights)
	edge_coords = coordinates['edges']
	print(len(edge_coords))
	lines = [ax.plot(xy[0], xy[1], alpha=0.9, linewidth=w, linestyle='-', color='#1a75ff', solid_capstyle='round')[0] for xy,w in zip(coordinates['edges'], line_weights)]


	ax.scatter(coordinates['x_f']+coordinates['x_e'], coordinates['y_f']+coordinates['y_e'],
					s=30, c='white', marker='o', lw=0, alpha=1)

	plot_words(ax, coordinates['x_f'], coordinates['y_f'], coordinates['w_f'], 'top')
	plot_words(ax, coordinates['x_e'], coordinates['y_e'], coordinates['w_e'], 'bottom')
	
	plt.savefig(fig_path + '.pdf')
	plt.clf()

def draw_alignment(naacl_path, french_path, english_path, fig_path, sure=False, sentence=1):
	"""

	"""
	i = 1
	f = open(french_path, 'r')
	while i <= sentence:
		french = f.readline().split()
		i += 1
	f.close()
	
	i = 1
	e = open(english_path, 'r')
	while i <= sentence:
		english = e.readline().split()
		i += 1
	e.close()

	n = open(naacl_path, 'r')
	alignments = defaultdict(list)
	line = n.readline()
	while line[0] != str(sentence):
		line = n.readline().split()
	while line[0] != str(sentence+1):
		if sure:
			if line[-1] != 'P':
				alignments[int(line[2])-1].append(int(line[1])-1)
		else:
			alignments[int(line[2])-1].append(int(line[1])-1)
		line = n.readline().split()
	
	bitext = ([(french, english)], alignments)

	fig = plt.figure(figsize=(8, 4))
	ax = plt.axes()
	plt.axis('off')
	coordinates = get_coordinates(bitext)

	
	lines = [ax.plot(xy[0], xy[1],alpha=0.9,linestyle='-',color='#1a75ff',linewidth=1.0, solid_capstyle='round')[0] for xy in coordinates['edges']]

	ax.scatter(coordinates['x_f']+coordinates['x_e'], coordinates['y_f']+coordinates['y_e'],
					s=30, c='white', marker='o', lw=0, alpha=1)

	plot_words(ax, coordinates['x_f'], coordinates['y_f'], coordinates['w_f'], 'top')
	plot_words(ax, coordinates['x_e'], coordinates['y_e'], coordinates['w_e'], 'bottom')
	
	plt.savefig(fig_path + '.pdf')



def draw_weighted_alignment(model, naacl_path, french_path, english_path, fig_path, sure=False, sentence=1):
	"""
	Draws an alignment that is weighted according to probs of alignment.
	"""
	i = 1
	f = open(french_path, 'r')
	while i <= sentence:
		french = f.readline().split()
		i += 1
	f.close()
	
	i = 1
	e = open(english_path, 'r')
	while i <= sentence:
		english = e.readline().split()
		i += 1
	e.close()

	n = open(naacl_path, 'r')
	alignments = defaultdict(list)
	line = n.readline()
	while line[0] != str(sentence):
		line = n.readline().split()
	while line[0] != str(sentence+1):
		if sure:
			if line[-1] != 'P':
				alignments[int(line[2])-1].append(int(line[1])-1)
		else:
			alignments[int(line[2])-1].append(int(line[1])-1)
		line = n.readline().split()
	
	bitext = ([(french, english)], alignments)

	fig = plt.figure(figsize=(8, 4))
	ax = plt.axes()
	plt.axis('off')
	coordinates = get_coordinates(bitext)

	# get weights of the edges:

	# F_indices = [model.V_f_indices[i] for i in french]
	# E_indices = [model.V_e_indices[i] for i in english]
	F_indices = []
	for i in french:
		try:
			f = model.V_f_indices[i]
		except KeyError:
			f = model.V_f_indices['-UNK-']
		F_indices.append(f)

	E_indices = []
	for i in english:
		try:
			e = model.V_e_indices[i]
		except KeyError:
			e = model.V_e_indices['-UNK-']
		F_indices.append(e)
	
	raw_line_weights = []
	
	for i, e in enumerate(E_indices):
		for f in F_indices:
			try:
				p = model.posterior(f, E_indices)[i]
			except IndexError:
				p = 0
			raw_line_weights.append(p)

	line_weights = [w*10 for w in raw_line_weights]
	edge_coords = coordinates['edges']
	lines = [ax.plot(xy[0], xy[1], alpha=0.9, linewidth=w, linestyle='-', color='#1a75ff', solid_capstyle='round')[0] for xy,w in zip(coordinates['edges'], line_weights)]


	ax.scatter(coordinates['x_f']+coordinates['x_e'], coordinates['y_f']+coordinates['y_e'],
					s=30, c='white', marker='o', lw=0, alpha=1)

	plot_words(ax, coordinates['x_f'], coordinates['y_f'], coordinates['w_f'], 'top')
	plot_words(ax, coordinates['x_e'], coordinates['y_e'], coordinates['w_e'], 'bottom')
	
	plt.savefig(fig_path)
	plt.clf()


def plot_words(axes, xs, ys, words, vertical_position, weight='bold', color='black'):
	"""
	Code borrowed from: https://github.com/INFR11133/lab1

	Plots a pair of sentences at given positions, on a given axes object.
	
	Args:
	   axes (pyplotlib axes object): axes on which to plot
	   xs (list of floats): x coordinates
	   ys (list of floats): y coordinates
	   words (list of strings): words to be displayed
	   vertical_position (string): where words should be displayed relative to point coordinates
	   weight (string): font weight
	   color (string or list of strings): color/s to be used for displaying words
	"""
	for n in range(0, len(words)):
		axes.text(xs[n], ys[n], words[n], size=9, family='sans-serif',
				  weight=weight, color=color,
				  horizontalalignment='center',
				  verticalalignment=vertical_position)


def get_coordinates(bitext, draw_all=False, one_sent=False, sent_index=0, word_index=0):
	"""
	Code borrowed from: https://github.com/INFR11133/lab1

	Generates x and y coordinates to be used for plotting sentence pairs
	and alignment links.
	
	Args:
	   bitext (list of tuples): list of translation pairs
	   one_sent (Boolean): whether coordinates ahould be generated for one
		   selected sentence pair and one word in it, or for the whole bitext
	   sent_index (int): index of the selected sentence pair
	   word_index (int): index of the target foreign word
	"""
	x_positions_f = []
	y_positions_f = []
	x_positions_e = []
	y_positions_e = []
	edge_pos = []
	words_f = []
	words_e = []
	sents, alignments = bitext
	for (n, (f, e)) in enumerate(sents):

		for j in range(0, len(f)):
			x_positions_f.append(j+1)
			y_positions_f.append((3*n)-2)
			words_f.append(f[j])
			if (not one_sent) or (one_sent and word_index==j):
				for i in range(0, len(e)):
					if draw_all:
						edge_pos.append([[j+1, i+1], [(3*n)-1.9, (3*n)-1.1]])
					else:
						if i in alignments[j]:
							edge_pos.append([[j+1, i+1], [(3*n)-1.9, (3*n)-1.1]])

		for i in range(0, len(e)):
			x_positions_e.append(i+1)
			y_positions_e.append((3*n)-1)
			words_e.append(e[i])
	coord_dict = {'x_f': x_positions_f, 'x_e': x_positions_e,
			'y_f': y_positions_f, 'y_e': y_positions_e,
			'edges': edge_pos, 'w_f': words_f, 'w_e': words_e}
	return coord_dict