def set_model(args, prepare_tab):
if args.blocks_dims == "same":
n_inp_dim = sum([e[2] for e in prepare_tab.embeddings_input])
blocks_dims = [n_inp_dim, n_inp_dim, n_inp_dim]
else:
blocks_dims = eval(args.blocks_dims)
if args.mlp_hidden_dims == "auto":
n_inp_dim = blocks_dims[-1]
mlp_hidden_dims = [4 * n_inp_dim, 2 * n_inp_dim]
else:
mlp_hidden_dims = eval(args.mlp_hidden_dims)
deeptabular = TabResnet(
embed_input=prepare_tab.embeddings_input,
column_idx=prepare_tab.column_idx,
blocks_dims=blocks_dims,
blocks_dropout=args.blocks_dropout,
mlp_hidden_dims=mlp_hidden_dims,
mlp_activation=args.mlp_activation,
mlp_dropout=args.mlp_dropout,
mlp_batchnorm=args.mlp_batchnorm,
mlp_batchnorm_last=args.mlp_batchnorm_last,
mlp_linear_first=args.mlp_linear_first,
embed_dropout=args.embed_dropout,
)
model = WideDeep(deeptabular=deeptabular)
return model
def test_full_setup(concat_cont_first):
model4 = TabResnet(
embed_input=embed_input,
column_idx={k: v
for v, k in enumerate(colnames)},
blocks_dims=[32, 16, 8],
blocks_dropout=0.5,
mlp_dropout=0.5,
mlp_hidden_dims=[32, 16],
mlp_batchnorm=True,
mlp_batchnorm_last=False,
embed_dropout=0.1,
continuous_cols=continuous_cols,
batchnorm_cont=True,
concat_cont_first=concat_cont_first,
)
out = model4(X_tab)
true_mlp_inp_dim = list(
model4.tab_resnet_mlp.mlp.dense_layer_0.parameters())[2].size(1)
if concat_cont_first:
expected_mlp_inp_dim = model4.blocks_dims[-1]
else:
expected_mlp_inp_dim = model4.blocks_dims[-1] + len(continuous_cols)
assert (out.size(0) == 10 and out.size(1) == model4.output_dim
and expected_mlp_inp_dim == true_mlp_inp_dim)
def test_cont_contat(concat_cont_first):
model3 = TabResnet(
blocks_dims=[32, 16, 8],
blocks_dropout=0.5,
mlp_dropout=0.5,
column_idx={k: v
for v, k in enumerate(colnames)},
embed_input=embed_input,
continuous_cols=continuous_cols,
concat_cont_first=concat_cont_first,
)
out = model3(X_tab)
assert out.size(0) == 10 and out.size(1) == model3.output_dim
blocks_dims = eval(args.blocks_dims)
if args.mlp_hidden_dims == "auto":
n_inp_dim = blocks_dims[-1]
mlp_hidden_dims = [4 * n_inp_dim, 2 * n_inp_dim]
else:
mlp_hidden_dims = eval(args.mlp_hidden_dims)
deeptabular = TabResnet(
embed_input=prepare_tab.embeddings_input,
column_idx=prepare_tab.column_idx,
blocks_dims=blocks_dims,
blocks_dropout=args.blocks_dropout,
mlp_hidden_dims=mlp_hidden_dims,
mlp_activation=args.mlp_activation,
mlp_dropout=args.mlp_dropout,
mlp_batchnorm=args.mlp_batchnorm,
mlp_batchnorm_last=args.mlp_batchnorm_last,
mlp_linear_first=args.mlp_linear_first,
embed_dropout=args.embed_dropout,
continuous_cols=prepare_tab.continuous_cols,
batchnorm_cont=args.batchnorm_cont,
concat_cont_first=args.concat_cont_first,
)
model = WideDeep(deeptabular=deeptabular)
optimizers = set_optimizer(model, args)
steps_per_epoch = (X_train.shape[0] // args.batch_size) + 1
lr_schedulers = set_lr_scheduler(optimizers, steps_per_epoch, args)
early_stopping = EarlyStopping(
monitor=args.monitor,
colnames = list(string.ascii_lowercase)[:10]
embed_cols = [np.random.choice(np.arange(5), 10) for _ in range(5)]
cont_cols = [np.random.rand(10) for _ in range(5)]
X_tab = torch.from_numpy(np.vstack(embed_cols + cont_cols).transpose())
X_tab_emb = X_tab[:, :5]
X_tab_cont = X_tab[:, 5:]
###############################################################################
# Embeddings and no continuous_cols
###############################################################################
embed_input = [(u, i, j) for u, i, j in zip(colnames[:5], [5] * 5, [16] * 5)]
model1 = TabResnet(
blocks_dims=[32, 16],
blocks_dropout=0.5,
mlp_dropout=0.5,
column_idx={k: v
for v, k in enumerate(colnames[:5])},
embed_input=embed_input,
)
def test_tab_resnet_embed():
out = model1(X_tab_emb)
assert out.size(0) == 10 and out.size(1) == 16
###############################################################################
# Continous Cols and Embeddings
###############################################################################
continuous_cols = colnames[-5:]
model2 = TabResnet(
) = train_test_split(X_wide, X_tab, X_text, X_img, target)
# build model components
wide = Wide(np.unique(X_wide).shape[0], 1)
tabmlp = TabMlp(
mlp_hidden_dims=[32, 16],
mlp_dropout=[0.5, 0.5],
column_idx={k: v
for v, k in enumerate(colnames)},
embed_input=embed_input,
continuous_cols=colnames[-5:],
)
tabresnet = TabResnet(
blocks_dims=[32, 16],
column_idx={k: v
for v, k in enumerate(colnames)},
embed_input=embed_input,
continuous_cols=colnames[-5:],
)
tabtransformer = TabTransformer(
column_idx={k: v
for v, k in enumerate(colnames)},
embed_input=embed_input_tt,
continuous_cols=colnames[5:],
)
deeptext = DeepText(vocab_size=vocab_size, embed_dim=32, padding_idx=0)
deepimage = DeepImage(pretrained=True)
###############################################################################
# test consistecy between optimizers and lr_schedulers format
###############################################################################