done

dev-0/out.tsv

@@ -0,0 +1,471 @@
+ja
+o
+?
+do
+mego
+:
+a
+co
+.
+po
+,
+kiedy
+!
+a
+ja
+o
+.
+i
+ja
+na
+jej
+do
+domu
+i
+do
+nas
+.
+ale
+co
+do
+nich
+!
+a
+!
+po
+,
+ażeby
+i
+po
+.
+i
+na
+jego
+do
+drzwi
+nie
+.
+a
+może
+o
+?
+?
+i
+co
+o
+niej
+i
+po
+południu
+:
+a
+jeżeli
+?
+)
+,
+ale
+i
+jeszcze
+do
+nas
+i
+co
+!
+nie
+miał
+do
+głowy
+;
+do
+niego
+.
+ale
+już
+już
+na
+mnie
+.
+po
+,
+ale
+nawet
+na
+nią
+:
+i
+o
+mnie
+,
+co
+pan
+do
+niej
+do
+niego
+?
+)
+?
+po
+?
+?
+i
+o
+niej
+!
+i
+do
+końca
+?
+po
+chwili
+.
+co
+i
+co
+o
+czym
+nie
+.
+a
+gdy
+.
+ja
+?
+)
+na
+jej
+i
+co
+na
+mnie
+;
+ja
+nie
+ma
+na
+jej
+;
+lecz
+co
+,
+co
+.
+pan
+co
+do
+pani
+i
+nawet
+,
+czy
+na
+mnie
+:
+na
+kilka
+?
+a
+co
+do
+mnie
+,
+a
+,
+o
+co
+na
+drugi
+:
+po
+?
+?
+o
+jej
+.
+ale
+i
+po
+?
+.
+i
+ja
+,
+czy
+na
+mnie
+nie
+będzie
+i
+jeszcze
+na
+jej
+na
+co
+dzień
+,
+na
+drugi
+na
+niego
+?
+po
+.
+pani
+i
+.
+na
+jej
+na
+nią
+i
+po
+,
+co
+,
+na
+mnie
+do
+okna
+,
+czy
+,
+i
+o
+jego
+na
+kilka
+nie
+mam
+na
+mnie
+i
+o
+?
+na
+lewo
+nie
+będzie
+.
+i
+ja
+i
+nawet
+?
+.
+po
+,
+ażeby
+mu
+do
+niego
+o
+,
+który
+ma
+na
+drugi
+na
+jej
+na
+co
+dzień
+.
+gdy
+już
+,
+bo
+do
+panny
+.
+a
+nie
+na
+jej
+na
+drugi
+?
+,
+czy
+i
+co
+nie
+jest
+i
+nie
+mam
+na
+jej
+na
+myśl
+?
+i
+ja
+i
+ja
+i
+po
+co
+.
+i
+na
+nią
+o
+nim
+!
+na
+jego
+nie
+jest
+nie
+,
+i
+już
+o
+jej
+do
+nas
+i
+co
+?
+o
+.
+na
+lewo
+,
+co
+mu
+nie
+mam
+i
+ja
+o
+jej
+nie
+ma
+?
+nie
+mam
+na
+kilka
+i
+ja
+o
+.
+ale
+co
+na
+jego
+i
+nawet
+i
+nie
+mam
+,
+a
+ja
+do
+nas
+nie
+miał
+.
+pani
+nie
+jest
+?
+.
+gdy
+na
+co
+na
+co
+.
+po
+ja
+,
+czy
+nie
+może
+na
+co
+.
+a
+na
+mnie
+nie
+ma
+,
+czy
+nie
+będzie
+,
+a
+nawet
+i
+na
+jego
+?
+na
+co
+,
+a
+nie
+ma
+do
+niej
+o
+mnie
+na
+.
+co
+i
+pani
+meliton
+:
+co
+?
+i
+po
+południu
+do
+niego
+.
+po
+co
+?
+i
+nawet
+!
+nie
+jest
+mnie
+!
+na
+mnie
+?
+czy
+do
+niego
+na
+jej
+;
+ale
+na
+co
+!
+o
+nie
+jest
+mnie
+do
+mego
+!
+na
+mnie
+?
+czy
+ja
+.
+na
+lewo
+do
+drzwi

script.py

@@ -0,0 +1,202 @@
+import numpy as np
+import torch
+from sklearn.model_selection import train_test_split
+import nltk
+from nltk.tokenize import word_tokenize
+
+nltk.download('punkt')
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+train_a = "train/train.tsv"
+lalka_path_train = 'train/train_train.tsv'
+lalka_path_valid = 'train/train_test.tsv'
+corpora_train = open(lalka_path_train).read()
+corpora_train_tokenized = list(word_tokenize(corpora_train))
+corpora_train_tokenized = [token.lower() for token in corpora_train_tokenized]
+
+vocab_itos = sorted(set(corpora_train_tokenized))
+vocab_itos = vocab_itos[:15005]
+vocab_itos[15001] = "<UNK>"
+vocab_itos[15002] = "<BOS>"
+vocab_itos[15003] = "<EOS>"
+vocab_itos[15004] = "<PAD>"
+
+BATCH_SIZE = 128
+EPOCHS = 15
+
+
+history_ppl_train = []
+history_ppl_valid = []
+
+vocab_stoi = dict()
+for i, token in enumerate(vocab_itos):
+    vocab_stoi[token] = i
+
+NGRAMS = 5
+
+
+def set_ppl(dataset_id_list):
+    lm.eval()
+
+    batches = 0
+    loss_sum = 0
+
+    for i in range(0, len(dataset_id_list) - BATCH_SIZE + 1, BATCH_SIZE):
+        X = dataset_id_list[i:i + BATCH_SIZE, :NGRAMS - 1]
+        Y = dataset_id_list[i:i + BATCH_SIZE, NGRAMS - 1]
+        predictions = lm(X)
+
+        loss = criterion(predictions, Y)
+
+        loss_sum += loss.item()
+        batches += 1
+
+    return np.exp(loss_sum / batches)
+
+
+def open_files(path_a, path_b, path_c):
+    with open(path_a, "r") as path:
+        lines = path.readlines()
+    train, test = train_test_split(lines, test_size=0.2)
+    with open(path_b, "w") as out_train_file:
+        for i in train:
+            out_train_file.write(i)
+    with open(path_c, "w") as out_test_file:
+        for i in test:
+            out_test_file.write(i)
+
+
+def get_samples(dataset):
+    samples = []
+    for i in range(len(dataset) - NGRAMS):
+        samples.append(dataset[i:i + NGRAMS])
+    return samples
+
+
+def get_token_id(dataset):
+    token_id_list = [vocab_stoi['<PAD>']] * (NGRAMS - 1) + [vocab_stoi['<BOS>']]
+    for token in dataset:
+        try:
+            token_id_list.append(vocab_stoi[token])
+        except KeyError:
+            token_id_list.append(vocab_stoi['<UNK>'])
+    token_id_list.append(vocab_stoi['<EOS>'])
+    return token_id_list
+
+
+open_files(train_a, lalka_path_train, lalka_path_valid)
+train_id_list = get_token_id(corpora_train_tokenized)
+train_id_list = get_samples(train_id_list)
+train_id_list = torch.tensor(train_id_list, device=device)
+corpora_valid = open(lalka_path_valid).read()
+corpora_valid_tokenized = list(word_tokenize(corpora_valid))
+corpora_valid_tokenized = [token.lower() for token in corpora_valid_tokenized]
+valid_id_list = get_token_id(corpora_valid_tokenized)
+valid_id_list = torch.tensor(get_samples(valid_id_list), dtype=torch.long, device=device)
+
+
+class GRU(torch.nn.Module):
+
+    def __init__(self):
+        super(GRU, self).__init__()
+        self.emb = torch.nn.Embedding(len(vocab_itos), 100)
+        self.rec = torch.nn.GRU(100, 256, 1, batch_first=True)
+        self.fc1 = torch.nn.Linear(256, len(vocab_itos))
+
+    def forward(self, x):
+        emb = self.emb(x)
+        output, h_n = self.rec(emb)
+        hidden = h_n.squeeze(0)
+        out = self.fc1(hidden)
+        return out
+
+
+lm = GRU().to(device)
+criterion = torch.nn.CrossEntropyLoss()
+optimizer = torch.optim.Adam(lm.parameters(), lr=0.0001)
+
+for epoch in range(EPOCHS):
+
+    batches = 0
+    loss_sum = 0
+    acc_score = 0
+    lm.train()
+    for i in range(0, len(train_id_list) - BATCH_SIZE + 1, BATCH_SIZE):
+        X = train_id_list[i:i + BATCH_SIZE, :NGRAMS - 1]
+        Y = train_id_list[i:i + BATCH_SIZE, NGRAMS - 1]
+        predictions = lm(X)
+        loss = criterion(predictions, Y)
+
+        optimizer.zero_grad()
+        loss.backward()
+        optimizer.step()
+
+        loss_sum += loss.item()
+        batches += 1
+
+    ppl_train = set_ppl(train_id_list)
+    ppl_valid = set_ppl(valid_id_list)
+
+    history_ppl_train.append(ppl_train)
+    history_ppl_valid.append(ppl_valid)
+
+
+tokenized = list(word_tokenize('Gości innych nie widział oprócz spółleśników'))
+tokenized = [token.lower() for token in tokenized]
+
+id_list = []
+for word in tokenized:
+    if word in vocab_stoi:
+        id_list.append(vocab_stoi[word])
+    else:
+        id_list.append(vocab_stoi['<UNK>'])
+
+lm.eval()
+
+id_list = torch.tensor(id_list, dtype=torch.long, device=device)
+
+preds = lm(id_list.unsqueeze(0))
+
+vocab_itos[torch.argmax(torch.softmax(preds, 1), 1).item()]
+
+tokenized = list(word_tokenize('Lalka'))
+tokenized = [token.lower() for token in tokenized]
+
+id_list = []
+for word in tokenized:
+    if word in vocab_stoi:
+        id_list.append(vocab_stoi[word])
+    else:
+        id_list.append(vocab_stoi['<UNK>'])
+id_list = torch.tensor([id_list], dtype=torch.long, device=device)
+
+candidates_number = 10
+for i in range(30):
+    preds = lm(id_list)
+    candidates = torch.topk(torch.softmax(preds, 1), candidates_number)[1][0].cpu().numpy()
+    candidate = 15001
+    while candidate > 15000:
+        candidate = candidates[np.random.randint(candidates_number)]
+    id_list = torch.cat((id_list, torch.tensor([[candidate]], device=device)), 1)
+
+with open("dev-0/in.tsv", "r") as dev_path:
+    nr_of_dev_lines = len(dev_path.readlines())
+with open("test-A/in.tsv", "r") as test_a_path:
+    nr_of_test_a_lines = len(test_a_path.readlines())
+with open("dev-0/out.tsv", "w") as out_dev_file:
+    for i in range(nr_of_dev_lines):
+        preds = lm(id_list)
+        candidates = torch.topk(torch.softmax(preds, 1), candidates_number)[1][0].cpu().numpy()
+        candidate = 15001
+        while candidate > 15000:
+            candidate = candidates[np.random.randint(candidates_number)]
+        id_list = torch.cat((id_list, torch.tensor([[candidate]], device=device)), 1)
+        out_dev_file.write(vocab_itos[candidate] + '\n')
+with open("test-A/out.tsv", "w") as out_test_file:
+    for i in range(nr_of_dev_lines):
+        preds = lm(id_list)
+        candidates = torch.topk(torch.softmax(preds, 1), candidates_number)[1][0].cpu().numpy()
+        candidate = 15001
+        while candidate > 15000:
+            candidate = candidates[np.random.randint(candidates_number)]
+        id_list = torch.cat((id_list, torch.tensor([[candidate]], device=device)), 1)
+        out_test_file.write(vocab_itos[candidate] + '\n')

test-A/out.tsv

@@ -0,0 +1,471 @@
+?
+a
+może
+o
+jej
+.
+na
+kilka
+dni
+i
+po
+południu
+,
+ale
+ja
+o
+.
+po
+i
+po
+prostu
+na
+mnie
+.
+pani
+na
+jej
+,
+jak
+ja
+i
+ja
+?
+po
+południu
+nie
+miał
+do
+nas
+nie
+było
+go
+o
+nie
+mam
+.
+i
+co
+do
+mnie
+!
+a
+po
+i
+ja
+,
+na
+niego
+i
+co
+?
+,
+a
+i
+nawet
+:
+do
+nas
+,
+na
+jego
+nie
+był
+.
+a
+jeżeli
+do
+domu
+!
+nie
+było
+do
+pani
+meliton
+.
+gdy
+na
+kilka
+na
+jej
+?
+)
+,
+na
+kilka
+na
+jej
+,
+nie
+miał
+do
+czasu
+na
+jej
+!
+)
+.
+ale
+ja
+nie
+mam
+.
+pani
+na
+jej
+nie
+będzie
+i
+na
+jego
+do
+.
+pani
+nie
+był
+?
+nie
+,
+co
+pan
+!
+o
+jej
+.
+ale
+na
+mnie
+,
+który
+?
+)
+o
+na
+jego
+i
+po
+kilka
+?
+.
+gdy
+,
+i
+na
+co
+o
+,
+bo
+.
+i
+już
+o
+nim
+do
+nas
+:
+na
+mnie
+i
+nawet
+nie
+było
+go
+.
+i
+gdy
+na
+mnie
+na
+kilka
+dni
+.
+po
+co
+do
+niej
+:
+o
+,
+i
+na
+nią
+:
+czy
+pan
+ignacy
+!
+)
+.
+ja
+,
+czy
+o
+nie
+był
+o
+czym
+do
+głowy
+!
+po
+co
+?
+o
+.
+po
+.
+na
+lewo
+.
+nie
+jestem
+,
+nie
+na
+nią
+o
+nim
+na
+co
+i
+nie
+na
+mnie
+do
+niej
+?
+)
+i
+o
+czym
+?
+nie
+ma
+pan
+o
+na
+kilka
+,
+ażeby
+,
+ażeby
+.
+ale
+już
+nie
+było
+do
+niego
+:
+po
+,
+a
+na
+co
+dzień
+i
+ja
+!
+)
+na
+jego
+dzień
+?
+a
+może
+o
+jego
+i
+na
+co
+do
+nas
+.
+nie
+,
+i
+o
+nie
+było
+.
+nie
+będzie
+mi
+po
+.
+co
+i
+po
+chwili
+na
+kilka
+.
+i
+o
+.
+i
+o
+na
+kilka
+na
+jego
+,
+który
+nie
+na
+drugi
+:
+co
+.
+ja
+.
+o
+jego
+nie
+ma
+do
+domu
+,
+bo
+na
+myśl
+:
+i
+o
+.
+po
+prostu
+,
+bo
+,
+a
+,
+i
+ja
+do
+czasu
+,
+i
+jeszcze
+do
+niego
+?
+na
+kilka
+:
+.
+ale
+co
+dzień
+,
+co
+pan
+na
+niego
+:
+i
+już
+do
+czasu
+,
+nie
+jest
+na
+niego
+,
+a
+jeżeli
+ja
+o
+jej
+.
+a
+może
+do
+niego
+do
+pani
+baronowej
+i
+nie
+jest
+,
+który
+do
+panny
+,
+ale
+ja
+o
+.
+a
+ja
+!
+ale
+o
+nie
+,
+nie
+miał
+do
+paryża
+!
+.
+i
+na
+jego
+do
+niego
+;
+po
+nie
+ma
+pani
+na
+nią
+.
+a
+ja
+.
+nie
+będzie
+o
+nie
+będzie
+do
+niego
+,
+ażeby
+na
+jego
+nie
+było
+go
+,
+bo
+już
+o
+czym
+;
+bo
+na
+nią
+do
+niej
+i
+nie
+na
+.
+pani
+?
+)
+,
+ale
+o
+na
+kilka
+.
+pani
+,
+co
+,