fix missing correct files

This commit is contained in:
kpierzynski 2024-05-22 02:48:52 +02:00
parent b13b78e58e
commit 1f224ccd28
4 changed files with 10624 additions and 10680 deletions

View File

@ -1,3 +1,3 @@
## challenging-america-word-gap-prediction
### using simple trigram nn
calculated perplexity: 583.35
calculated perplexity: 653.89

File diff suppressed because it is too large Load Diff

59
run.py
View File

@ -17,12 +17,12 @@ import torch
from tqdm.notebook import tqdm
embed_size = 100
vocab_size = 25_000
num_epochs = 1
embed_size = 30
vocab_size = 10_000
num_epochs = 2
device = 'cuda'
batch_size = 2048
train_file_path = 'train/train.txt'
batch_size = 8192
train_file_path = 'train/nano.txt'
with open(train_file_path, 'r', encoding='utf-8') as file:
total = len(file.readlines())
@ -43,7 +43,7 @@ def get_words_from_line(line):
def get_word_lines_from_file(file_name):
limit = total * 2
with open(file_name, 'r', encoding='utf8') as fh:
for line in tqdm(fh, total=total):
for line in fh:
limit -= 1
if not limit:
break
@ -82,7 +82,6 @@ train_dataset = Trigrams(train_file_path, vocab_size)
# In[3]:
# Neural network model for trigram language modeling
class SimpleTrigramNeuralLanguageModel(nn.Module):
def __init__(self, vocabulary_size, embedding_size):
super(SimpleTrigramNeuralLanguageModel, self).__init__()
@ -93,48 +92,50 @@ class SimpleTrigramNeuralLanguageModel(nn.Module):
self.embedding_size = embedding_size
def forward(self, x):
embeds = self.embedding(x).view(-1, self.embedding_size * 2)
out = torch.relu(self.linear1(embeds))
embeds = self.embedding(x).view(x.size(0), -1)
out = self.linear1(embeds)
out = self.linear2(out)
return self.softmax(out)
# Instantiate the model
model = SimpleTrigramNeuralLanguageModel(vocab_size, embed_size)
model = SimpleTrigramNeuralLanguageModel(vocab_size, embed_size).to(device)
# In[4]:
model = SimpleTrigramNeuralLanguageModel(vocab_size, embed_size).to(device)
data = DataLoader(train_dataset, batch_size=batch_size)
optimizer = torch.optim.Adam(model.parameters())
criterion = torch.nn.NLLLoss()
criterion = torch.nn.CrossEntropyLoss()
# In[5]:
model.train()
step = 0
for _ in range(num_epochs):
for x1,x2,y in tqdm(data, total=total):
x = torch.cat((x1,x2), dim=0).to(device)
for x1,x2,y in tqdm(data, desc="Train loop"):
y = y.to(device)
x = torch.cat((x1.unsqueeze(1),x2.unsqueeze(1)), dim=1).to(device)
optimizer.zero_grad()
ypredicted = model(x)
loss = criterion(torch.log(ypredicted), y)
if step % 5000 == 0:
print(step, loss)
step += 1
loss.backward()
optimizer.step()
step = 0
model.eval()
# In[10]:
# In[6]:
def get_gap_candidates(words, n=10, vocab=train_dataset.vocab):
ixs = vocab.forward(words)
ixs = torch.tensor(ixs)
ixs = torch.cat(tuple([ixs]), dim=0).to(device)
ixs = vocab(words)
ixs = torch.tensor(ixs).unsqueeze(0).to(device)
out = model(ixs)
top = torch.topk(out[0], n)
@ -144,7 +145,7 @@ def get_gap_candidates(words, n=10, vocab=train_dataset.vocab):
return list(zip(top_words, top_probs))
# In[11]:
# In[7]:
def clean(text):
@ -172,13 +173,7 @@ def predictor(prefix):
return output
# In[12]:
predictor("I really bug")
# In[13]:
# In[8]:
def generate_result(input_path, output_path='out.tsv'):
@ -191,14 +186,8 @@ def generate_result(input_path, output_path='out.tsv'):
output_file.write(result + '\n')
# In[14]:
# In[9]:
generate_result('dev-0/in.tsv', output_path='dev-0/out.tsv')
# In[ ]:

View File

@ -10,11 +10,11 @@
"name": "stderr",
"output_type": "stream",
"text": [
"S:\\WENV\\Lib\\site-packages\\torchtext\\vocab\\__init__.py:4: UserWarning: \n",
"S:\\WENV_TORCHTEXT\\Lib\\site-packages\\torchtext\\vocab\\__init__.py:4: UserWarning: \n",
"/!\\ IMPORTANT WARNING ABOUT TORCHTEXT STATUS /!\\ \n",
"Torchtext is deprecated and the last released version will be 0.18 (this one). You can silence this warning by calling the following at the beginnign of your scripts: `import torchtext; torchtext.disable_torchtext_deprecation_warning()`\n",
" warnings.warn(torchtext._TORCHTEXT_DEPRECATION_MSG)\n",
"S:\\WENV\\Lib\\site-packages\\torchtext\\utils.py:4: UserWarning: \n",
"S:\\WENV_TORCHTEXT\\Lib\\site-packages\\torchtext\\utils.py:4: UserWarning: \n",
"/!\\ IMPORTANT WARNING ABOUT TORCHTEXT STATUS /!\\ \n",
"Torchtext is deprecated and the last released version will be 0.18 (this one). You can silence this warning by calling the following at the beginnign of your scripts: `import torchtext; torchtext.disable_torchtext_deprecation_warning()`\n",
" warnings.warn(torchtext._TORCHTEXT_DEPRECATION_MSG)\n"
@ -35,12 +35,12 @@
"\n",
"from tqdm.notebook import tqdm\n",
"\n",
"embed_size = 100\n",
"vocab_size = 25_000\n",
"num_epochs = 1\n",
"embed_size = 30\n",
"vocab_size = 10_000\n",
"num_epochs = 2\n",
"device = 'cuda'\n",
"batch_size = 2048\n",
"train_file_path = 'train/train.txt'\n",
"batch_size = 8192\n",
"train_file_path = 'train/nano.txt'\n",
"\n",
"with open(train_file_path, 'r', encoding='utf-8') as file:\n",
" total = len(file.readlines())"
@ -51,22 +51,7 @@
"execution_count": 2,
"id": "40392665-79bc-4032-a5de-9d189545c9f7",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "5049d1e295954b7baf71ac05a793071a",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/432022 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"outputs": [],
"source": [
"# Function to extract words from a line of text\n",
"def get_words_from_line(line):\n",
@ -80,7 +65,7 @@
"def get_word_lines_from_file(file_name):\n",
" limit = total * 2\n",
" with open(file_name, 'r', encoding='utf8') as fh:\n",
" for line in tqdm(fh, total=total):\n",
" for line in fh:\n",
" limit -= 1\n",
" if not limit:\n",
" break\n",
@ -123,7 +108,6 @@
"metadata": {},
"outputs": [],
"source": [
"# Neural network model for trigram language modeling\n",
"class SimpleTrigramNeuralLanguageModel(nn.Module):\n",
" def __init__(self, vocabulary_size, embedding_size):\n",
" super(SimpleTrigramNeuralLanguageModel, self).__init__()\n",
@ -134,44 +118,41 @@
" self.embedding_size = embedding_size\n",
"\n",
" def forward(self, x):\n",
" embeds = self.embedding(x).view(-1, self.embedding_size * 2)\n",
" out = torch.relu(self.linear1(embeds))\n",
" embeds = self.embedding(x).view(x.size(0), -1)\n",
" out = self.linear1(embeds)\n",
" out = self.linear2(out)\n",
" return self.softmax(out)\n",
"\n",
"# Instantiate the model\n",
"model = SimpleTrigramNeuralLanguageModel(vocab_size, embed_size)"
"model = SimpleTrigramNeuralLanguageModel(vocab_size, embed_size).to(device)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "32ea22db-7259-4549-a9d5-4781d9bc99bc",
"metadata": {},
"outputs": [],
"source": [
"data = DataLoader(train_dataset, batch_size=batch_size)\n",
"optimizer = torch.optim.Adam(model.parameters())\n",
"criterion = torch.nn.CrossEntropyLoss()"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "0858967e-5143-4253-921d-a009dbbdca27",
"metadata": {},
"outputs": [
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "c422bb888518406e9f6a4a8f10f2b473",
"model_id": "5e4b6ce6edf94b90a70d415d75be7eb6",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/432022 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "4020a53c31544ef3b4017c43798aa305",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/432022 [00:00<?, ?it/s]"
"Train loop: 0it [00:00, ?it/s]"
]
},
"metadata": {},
@ -181,73 +162,76 @@
"name": "stdout",
"output_type": "stream",
"text": [
"0 tensor(10.1654, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"5000 tensor(6.5147, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"10000 tensor(6.6747, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"15000 tensor(6.9061, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"20000 tensor(6.8899, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"25000 tensor(6.8373, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"30000 tensor(6.8942, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"35000 tensor(6.9564, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"40000 tensor(6.9709, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"45000 tensor(6.9592, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"50000 tensor(6.8195, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"55000 tensor(6.7074, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"60000 tensor(6.8755, device='cuda:0', grad_fn=<NllLossBackward0>)\n",
"65000 tensor(6.9605, device='cuda:0', grad_fn=<NllLossBackward0>)\n"
"0 tensor(9.2450, device='cuda:0', grad_fn=<NllLossBackward0>)\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "044c2ea05e344306881002e34d89bd54",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
"Train loop: 0it [00:00, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"0 tensor(6.2669, device='cuda:0', grad_fn=<NllLossBackward0>)\n"
]
},
{
"data": {
"text/plain": [
"SimpleTrigramNeuralLanguageModel(\n",
" (embedding): Embedding(25000, 100)\n",
" (linear1): Linear(in_features=200, out_features=100, bias=True)\n",
" (linear2): Linear(in_features=100, out_features=25000, bias=True)\n",
" (embedding): Embedding(10000, 30)\n",
" (linear1): Linear(in_features=60, out_features=30, bias=True)\n",
" (linear2): Linear(in_features=30, out_features=10000, bias=True)\n",
" (softmax): Softmax(dim=1)\n",
")"
]
},
"execution_count": 4,
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"model = SimpleTrigramNeuralLanguageModel(vocab_size, embed_size).to(device)\n",
"data = DataLoader(train_dataset, batch_size=batch_size)\n",
"optimizer = torch.optim.Adam(model.parameters())\n",
"criterion = torch.nn.NLLLoss()\n",
"\n",
"model.train()\n",
"step = 0\n",
"for _ in range(num_epochs):\n",
" for x1,x2,y in tqdm(data, total=total):\n",
" x = torch.cat((x1,x2), dim=0).to(device)\n",
" for x1,x2,y in tqdm(data, desc=\"Train loop\"):\n",
" y = y.to(device)\n",
" x = torch.cat((x1.unsqueeze(1),x2.unsqueeze(1)), dim=1).to(device)\n",
" optimizer.zero_grad()\n",
" ypredicted = model(x)\n",
" \n",
" loss = criterion(torch.log(ypredicted), y)\n",
" if step % 5000 == 0:\n",
" print(step, loss)\n",
" step += 1\n",
" loss.backward()\n",
" optimizer.step()\n",
"\n",
" step = 0\n",
"model.eval()"
]
},
{
"cell_type": "code",
"execution_count": 10,
"execution_count": 6,
"id": "da4d116c-beec-436d-84d8-577282507226",
"metadata": {},
"outputs": [],
"source": [
"def get_gap_candidates(words, n=10, vocab=train_dataset.vocab):\n",
" ixs = vocab.forward(words)\n",
" ixs = torch.tensor(ixs)\n",
" ixs = torch.cat(tuple([ixs]), dim=0).to(device)\n",
" ixs = vocab(words)\n",
" ixs = torch.tensor(ixs).unsqueeze(0).to(device)\n",
"\n",
" out = model(ixs)\n",
" top = torch.topk(out[0], n)\n",
@ -259,7 +243,7 @@
},
{
"cell_type": "code",
"execution_count": 11,
"execution_count": 7,
"id": "0cafd70a-29b3-4a49-b40f-b8ce3143084a",
"metadata": {},
"outputs": [],
@ -291,28 +275,7 @@
},
{
"cell_type": "code",
"execution_count": 12,
"id": "2084fb5f-6405-4e44-a06f-953db852e526",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'the:0.07267232984304428 of:0.043321046978235245 and:0.032147664576768875 to:0.02692588046193123 a:0.020654045045375824 in:0.020213929936289787 that:0.010836434550583363 is:0.00959325022995472 it:0.008407277055084705 :0.755228141322732'"
]
},
"execution_count": 12,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"predictor(\"I really bug\")"
]
},
{
"cell_type": "code",
"execution_count": 13,
"execution_count": 8,
"id": "965ebaf3-4c0b-4462-8ac5-4746ec9489ab",
"metadata": {},
"outputs": [],
@ -329,21 +292,13 @@
},
{
"cell_type": "code",
"execution_count": 14,
"execution_count": 9,
"id": "80547ba7-9d01-4d2b-9e83-269919513de9",
"metadata": {},
"outputs": [],
"source": [
"generate_result('dev-0/in.tsv', output_path='dev-0/out.tsv')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1d7d72b0-d629-487e-bcec-4756fa88ae49",
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {