Skip to content

Instantly share code, notes, and snippets.

import gc
def objective(trial):
### Dataset section
# see https://gist.github.com/krsnewwave/273b9cafa4813771791f076cee32c2e4#file-nvtabular_movielens_main_loop_functions-py-L2
train_loader, valid_loader = create_loaders(train_dataset, valid_dataset)
### Model section
# see https://gist.github.com/krsnewwave/273b9cafa4813771791f076cee32c2e4#file-nvtabular_movielens_main_loop_functions-py-L29
epochs = 1
patience = 3
# (1) Create loaders
def create_loaders(train_dataset, valid_dataset):
# dataset and loaders
train_iter = TorchAsyncItr(
train_dataset,
batch_size=BATCH_SIZE,
cats=CATEGORICAL_COLUMNS + CATEGORICAL_MH_COLUMNS,
conts=NUMERIC_COLUMNS,
labels=["rating"],
)
class WideAndDeepMultihot(pl.LightningModule):
# others go here...
#
#
def training_step(self, batch, batch_idx):
# unpack
x_cat, x_cont, y = self.transform.transform_with_label(batch)
# forward
y_pred = self((x_cat, x_cont))
import pytorch_lightning as pl
from nvtabular.framework_utils.torch.layers import ConcatenatedEmbeddings, MultiHotEmbeddings
import torch
class WideAndDeepMultihot(pl.LightningModule):
def __init__(
self,
model_conf,
cat_names, cont_names, label_names,
num_continuous,
@krsnewwave
krsnewwave / vae_pytorch.py
Last active May 23, 2022 14:57
vae recommender implementation on pytorch lightning
class MVAERecommender(TopNRecommender):
# TopNRecommender contains methods to predict top k
def __init__(self, model_conf : Dict, novelty_per_item, num_users, num_items, remove_observed = False, ):
# ... configuration is skipped
# # # # Model Structure # # # #
# this is to handle encoding dimensions as lists
self.encoder = nn.ModuleList()
# this enumeration produces dims in pairs, start with 1
for i, (d_in, d_out) in enumerate(zip(self.enc_dims[:-1], self.enc_dims[1:]), start=1):
# double d out at last for the mean and variance parameters
# pip install ray tune, and comet ml
from ray.tune.integration.comet import CometLoggerCallback
from functools import partial
from ray.tune.integration.pytorch_lightning import TuneReportCallback
def train_function(model_conf, novelty_per_item, epochs, patience,
train_loader, val_loader, checkpoint_dir=None):
model = CDAE(model_conf, novelty_per_item, num_users, num_items)
# fill up your metrics here
class CDAE(pl.LightningModule):
def __init__(self, model_conf : Dict, novelty_per_item, num_users, num_items, remove_observed = False, ):
super().__init__()
self.hidden_dim = model_conf["hidden_dim"]
# ... other self. initializations
self.user_embedding = nn.Embedding(self.num_users, self.hidden_dim)
self.encoder = nn.Linear(self.num_items, self.hidden_dim)
self.decoder = nn.Linear(self.hidden_dim, self.num_items)
class RecoSparseTrainDataset(Dataset):
def __init__(self, sparse_mat):
self.sparse_mat = sparse_mat
def __len__(self):
return self.sparse_mat.shape[0]
def __getitem__(self, idx):
batch_matrix = self.sparse_mat[idx].toarray().squeeze()
return batch_matrix, idx
class LightningResNet(pl.LightningModule):
def __init__(self, net_pretrained, device='cpu', criterion = F.cross_entropy,
num_classes = 4, optimizer = None, scheduler = None):
super().__init__()
self.net = net_pretrained
# set top to number of classes
num_ftrs = self.net.fc.in_features
self.net.fc = nn.Linear(num_ftrs, num_classes)
@krsnewwave
krsnewwave / pytorch_multitask_paintings.py
Created April 9, 2022 16:42
PyTorch model for multitask learning
# following https://towardsdatascience.com/multilabel-classification-with-pytorch-in-5-minutes-a4fa8993cbc7
class LightningResNetMultiLabel(pl.LightningModule):
def __init__(self, net, n_period, n_artists, criterion = F.cross_entropy, optimizer = None, scheduler = None, dropout_p = 0., lr=0.001, freeze_net=False):
super().__init__()
self.net = net
self.feature_extractor = nn.Sequential(*(list(self.net.children())[:-1]))
if freeze_net:
for param in self.net.parameters():