def __init__(self, model_pars=None, data_pars=None, compute_pars=None):
self.model_pars, self.compute_pars, self.data_pars = model_pars, compute_pars, data_pars
if model_pars is None:
self.model = None
else:
###############################################################
dm = data_pars['cols_model_group_custom']
data_config = DataConfig(
target=dm[
'coly'], #target should always be a list. Multi-targets are only supported for regression. Multi-Task Classification is not implemented
continuous_cols=dm['colnum'],
categorical_cols=dm['colcat'],
)
model_config = CategoryEmbeddingModelConfig(
**model_pars['model_pars'], )
trainer_config = TrainerConfig(
**compute_pars.get('compute_pars', {}))
optimizer_config = OptimizerConfig()
self.config_pars = {
'data_config': data_config,
'model_config': model_config,
'optimizer_config': optimizer_config,
'trainer_config': trainer_config,
}
self.model = TabularModel(**self.config_pars)
self.guide = None
self.pred_summary = None ### All MC summary
if VERBOSE: log(self.guide, self.model)
def test2(nrows=10000):
"""
python source/models/torch_tabular.py test
"""
global model, session
df,colcat, colnum, coly = test_dataset_covtype(1000)
target_name = coly
df.head()
train, test = train_test_split(df, random_state=42)
train, val = train_test_split(train, random_state=42)
num_classes = len(set(train[target_name].values.ravel()))
data_config = DataConfig(
target=target_name,
continuous_cols=colnum,
categorical_cols=colcat,
continuous_feature_transform=None,#"quantile_normal",
normalize_continuous_features=False
)
model_config = CategoryEmbeddingModelConfig(task="classification",
metrics=["f1","accuracy"],
metrics_params=[{"num_classes":num_classes},{}])
trainer_config = TrainerConfig(gpus=None, fast_dev_run=True)
experiment_config = ExperimentConfig(project_name="PyTorch Tabular Example",
run_name="node_forest_cov",
exp_watch="gradients",
log_target="wandb",
log_logits=True)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
# experiment_config=experiment_config,
)
tabular_model.fit( train=train, validation=val)
result = tabular_model.evaluate(val)
log(result)
test.drop(columns=target_name, inplace=True)
pred_df = tabular_model.predict(val.iloc[:100,:])
log(pred_df)
def test_regression(
regression_data,
multi_target,
continuous_cols,
categorical_cols,
continuous_feature_transform,
normalize_continuous_features,
target_range,
):
(train, test, target) = regression_data
if len(continuous_cols) + len(categorical_cols) == 0:
assert True
else:
data_config = DataConfig(
target=target + ["MedInc"] if multi_target else target,
continuous_cols=continuous_cols,
categorical_cols=categorical_cols,
continuous_feature_transform=continuous_feature_transform,
normalize_continuous_features=normalize_continuous_features,
)
model_config_params = dict(task="regression")
if target_range:
_target_range = []
for target in data_config.target:
_target_range.append((
train[target].min().item(),
train[target].max().item(),
))
model_config_params["target_range"] = _target_range
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(max_epochs=3,
checkpoints=None,
early_stopping=None,
gpus=0)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
tabular_model.fit(train=train, test=test)
result = tabular_model.evaluate(test)
# print(result[0]["valid_loss"])
assert "valid_loss" in result[0].keys()
pred_df = tabular_model.predict(test)
assert pred_df.shape[0] == test.shape[0]
def test_date_encoding(timeseries_data, freq):
(train, test, target) = timeseries_data
train, valid = train_test_split(train, random_state=42)
data_config = DataConfig(
target=target + ["Occupancy"],
continuous_cols=[
"Temperature", "Humidity", "Light", "CO2", "HumidityRatio"
],
categorical_cols=[],
date_columns=[("date", freq)],
encode_date_columns=True,
)
model_config_params = dict(task="regression")
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(max_epochs=1,
checkpoints=None,
early_stopping=None)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
config = tabular_model.config
datamodule = TabularDatamodule(
train=train,
validation=valid,
config=config,
test=test,
)
datamodule.prepare_data()
if freq != "S":
datamodule.setup("fit")
config = datamodule.config
if freq == "H":
assert "_Hour" in datamodule.train.columns
elif freq == "D":
assert "_Dayofyear" in datamodule.train.columns
elif freq == "T":
assert "_Minute" in datamodule.train.columns
elif freq == "S":
try:
datamodule.setup("fit")
assert False
except RuntimeError:
assert True
def test_embedding_transformer(regression_data):
(train, test, target) = regression_data
data_config = DataConfig(
target=target,
continuous_cols=[
"AveRooms",
"AveBedrms",
"Population",
"AveOccup",
"Latitude",
"Longitude",
],
categorical_cols=["HouseAgeBin"],
)
model_config_params = dict(task="regression")
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(
max_epochs=1,
checkpoints=None,
early_stopping=None,
gpus=None,
fast_dev_run=True,
)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
tabular_model.fit(train=train, test=test)
transformer = CategoricalEmbeddingTransformer(tabular_model)
train_transform = transformer.fit_transform(train)
embed_cols = [
col for col in train_transform.columns
if "HouseAgeBin_embed_dim" in col
]
assert len(train["HouseAgeBin"].unique()) + 1 == len(
transformer._mapping["HouseAgeBin"].keys())
assert all([
val.shape[0] == len(embed_cols)
for val in transformer._mapping["HouseAgeBin"].values()
])
def test_classification(
classification_data,
continuous_cols,
categorical_cols,
continuous_feature_transform,
normalize_continuous_features,
):
(train, test, target) = classification_data
if len(continuous_cols) + len(categorical_cols) == 0:
assert True
else:
data_config = DataConfig(
target=target,
continuous_cols=continuous_cols,
categorical_cols=categorical_cols,
continuous_feature_transform=continuous_feature_transform,
normalize_continuous_features=normalize_continuous_features,
)
model_config_params = dict(task="classification")
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(
max_epochs=3,
checkpoints=None,
early_stopping=None,
gpus=None,
fast_dev_run=True,
)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
tabular_model.fit(train=train, test=test)
result = tabular_model.evaluate(test)
# print(result[0]["valid_loss"])
assert "test_accuracy" in result[0].keys()
pred_df = tabular_model.predict(test)
assert pred_df.shape[0] == test.shape[0]
def main():
# Generate Synthetic Data
data, cat_col_names, num_col_names = make_mixed_classification(
n_samples=10000, n_features=20, n_categories=4)
train, test = train_test_split(data, random_state=42)
train, val = train_test_split(train, random_state=42)
# ##########Define the Configs############
data_config = DataConfig(target=["target"],
continuous_cols=num_col_names,
categorical_cols=cat_col_names)
trainer_config = TrainerConfig(auto_lr_find=True,
batch_size=1024,
max_epochs=100,
gpus=1)
optimizer_config = OptimizerConfig()
model_config = CategoryEmbeddingModelConfig(task="classification",
layers="1024-512-512",
activation="LeakyReLU",
learning_rate=1e-3)
tabular_mode = TabularModel(data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config)
# Training the Model
tabular_mode.fit(train=train, validation=val)
# Evaluating the Model
# #Loss and Metrics on New Data¶
result = tabular_mode.evaluate(test)
# #New Predictions as DataFrame
pred_df = tabular_mode.predict(test)
pred_df.head()
print_metrics(test['target'], pred_df["prediction"], tag="Holdout")
# saving model
tabular_mode.save_model("Analysis/basic")
def test_ssl(
regression_data,
continuous_cols,
categorical_cols,
continuous_feature_transform,
normalize_continuous_features,
target_range,
target_transform,
custom_metrics,
custom_loss,
custom_optimizer,
ssl_task,
aug_task,
):
(train, test, target) = regression_data
if len(continuous_cols) + len(categorical_cols) == 0:
assert True
else:
data_config = DataConfig(
target=target,
continuous_cols=continuous_cols,
categorical_cols=categorical_cols,
continuous_feature_transform=continuous_feature_transform,
normalize_continuous_features=normalize_continuous_features,
)
model_config_params = dict(task="ssl",
ssl_task=ssl_task,
aug_task=aug_task)
if target_range:
_target_range = []
for target in data_config.target:
_target_range.append((
float(train[target].min()),
float(train[target].max()),
))
model_config_params["target_range"] = _target_range
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(
max_epochs=3,
checkpoints=None,
early_stopping=None,
gpus=None,
fast_dev_run=True,
)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
tabular_model.fit(
train=train,
test=test,
metrics=custom_metrics,
target_transform=target_transform,
loss=custom_loss,
optimizer=custom_optimizer,
optimizer_params={},
)
result = tabular_model.evaluate(test)
if custom_metrics is None:
assert "test_mean_squared_error" in result[0].keys()
else:
assert "test_fake_metric" in result[0].keys()
def test_dataloader(
regression_data,
validation_split,
multi_target,
continuous_cols,
categorical_cols,
continuous_feature_transform,
normalize_continuous_features,
target_transform,
embedding_dims,
):
(train, test, target) = regression_data
train, valid = train_test_split(train, random_state=42)
if len(continuous_cols) + len(categorical_cols) == 0:
assert True
else:
data_config = DataConfig(
target=target + ["MedInc"] if multi_target else target,
continuous_cols=continuous_cols,
categorical_cols=categorical_cols,
continuous_feature_transform=continuous_feature_transform,
normalize_continuous_features=normalize_continuous_features,
validation_split=validation_split,
)
model_config_params = dict(task="regression",
embedding_dims=embedding_dims)
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(max_epochs=1,
checkpoints=None,
early_stopping=None)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
config = tabular_model.config
datamodule = TabularDatamodule(
train=train,
validation=valid,
config=config,
test=test,
target_transform=target_transform,
)
datamodule.prepare_data()
datamodule.setup("fit")
config = datamodule.config
if len(categorical_cols) > 0:
assert config.categorical_cardinality[0] == 5
if embedding_dims is None:
assert config.embedding_dims[0][-1] == 3
else:
assert config.embedding_dims[0][-1] == embedding_dims[0][-1]
if normalize_continuous_features and len(continuous_cols) > 0:
assert round(
datamodule.train[config.continuous_cols[0]].mean()) == 0
assert round(
datamodule.train[config.continuous_cols[0]].std()) == 1
# assert round(datamodule.validation[config.continuous_cols[0]].mean()) == 0
# assert round(datamodule.validation[config.continuous_cols[0]].std()) == 1
val_loader = datamodule.val_dataloader()
_val_loader = datamodule.prepare_inference_dataloader(valid)
chk_1 = next(iter(val_loader))["continuous"]
chk_2 = next(iter(_val_loader))["continuous"]
assert np.not_equal(chk_1, chk_2).sum().item() == 0
'target'
], #target should always be a list. Multi-targets are only supported for regression. Multi-Task Classification is not implemented
continuous_cols=num_col_names,
categorical_cols=cat_col_names,
)
trainer_config = TrainerConfig(
auto_lr_find=
True, # Runs the LRFinder to automatically derive a learning rate
batch_size=1024,
max_epochs=100,
gpus=1, #index of the GPU to use. 0, means CPU
)
optimizer_config = OptimizerConfig()
model_config = CategoryEmbeddingModelConfig(
task="classification",
layers="1024-512-512", # Number of nodes in each layer
activation="LeakyReLU", # Activation between each layers
learning_rate=1e-3)
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
tabular_model.fit(train=train, validation=val)
result = tabular_model.evaluate(test)
pred_df = tabular_model.predict(test)
tabular_model.save_model("Analysis/basic")
loaded_model = TabularModel.load_from_checkpoint("Analysis/basic")
def test2(nrows=10000):
"""
python source/models/torch_tabular.py test
"""
global model, session
#X = np.random.rand(10000,20)
#y = np.random.binomial(n=1, p=0.5, size=[10000])
BASE_DIR = Path.home().joinpath('data/input/covtype/')
datafile = BASE_DIR.joinpath('covtype.data.gz')
datafile.parent.mkdir(parents=True, exist_ok=True)
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/covtype/covtype.data.gz"
if not datafile.exists():
wget.download(url, datafile.as_posix())
target_name = ["Covertype"]
colcat = [
"Wilderness_Area1", "Wilderness_Area2", "Wilderness_Area3",
"Wilderness_Area4", "Soil_Type1", "Soil_Type2", "Soil_Type3",
"Soil_Type4", "Soil_Type5", "Soil_Type6", "Soil_Type7", "Soil_Type8",
"Soil_Type9", "Soil_Type10", "Soil_Type11", "Soil_Type12",
"Soil_Type13", "Soil_Type14", "Soil_Type15", "Soil_Type16",
"Soil_Type17", "Soil_Type18", "Soil_Type19", "Soil_Type20",
"Soil_Type21", "Soil_Type22", "Soil_Type23", "Soil_Type24",
"Soil_Type25", "Soil_Type26", "Soil_Type27", "Soil_Type28",
"Soil_Type29", "Soil_Type30", "Soil_Type31", "Soil_Type32",
"Soil_Type33", "Soil_Type34", "Soil_Type35", "Soil_Type36",
"Soil_Type37", "Soil_Type38", "Soil_Type39", "Soil_Type40"
]
colnum = [
"Elevation", "Aspect", "Slope", "Horizontal_Distance_To_Hydrology",
"Vertical_Distance_To_Hydrology", "Horizontal_Distance_To_Roadways",
"Hillshade_9am", "Hillshade_Noon", "Hillshade_3pm",
"Horizontal_Distance_To_Fire_Points"
]
feature_columns = (colnum + colcat + target_name)
df = pd.read_csv(datafile, header=None, names=feature_columns, nrows=nrows)
df.head()
train, test = train_test_split(df, random_state=42)
train, val = train_test_split(train, random_state=42)
num_classes = len(set(train[target_name].values.ravel()))
data_config = DataConfig(
target=target_name,
continuous_cols=colnum,
categorical_cols=colcat,
continuous_feature_transform=None, #"quantile_normal",
normalize_continuous_features=False)
model_config = CategoryEmbeddingModelConfig(task="classification",
metrics=["f1", "accuracy"],
metrics_params=[{
"num_classes":
num_classes
}, {}])
trainer_config = TrainerConfig(gpus=None, fast_dev_run=True)
experiment_config = ExperimentConfig(
project_name="PyTorch Tabular Example",
run_name="node_forest_cov",
exp_watch="gradients",
log_target="wandb",
log_logits=True)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
# experiment_config=experiment_config,
)
tabular_model.fit(train=train, validation=val)
result = tabular_model.evaluate(val)
log(result)
test.drop(columns=target_name, inplace=True)
pred_df = tabular_model.predict(val.iloc[:100, :])
log(pred_df)
categorical_cols=cat_col_names,
continuous_feature_transform="quantile_normal",
normalize_continuous_features=True
)
trainer_config = TrainerConfig(
auto_lr_find=True, # Runs the LRFinder to automatically derive a learning rate
batch_size=1024,
max_epochs=10,
gpus=1, #index of the GPU to use. 0, means CPU
fast_dev_run=True
)
optimizer_config = OptimizerConfig()
model_config = CategoryEmbeddingModelConfig(
task="classification",
layers="4096-4096-512", # Number of nodes in each layer
activation="LeakyReLU", # Activation between each layers
learning_rate = 1e-3,
metrics=["accuracy"]
)
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
tabular_model.fit(train=train, validation=val)
test.drop(columns=['target'], inplace=True)
pred_df = tabular_model.predict(test)
# tabular_model.fit(train=train, validation=val)
def test_date_encoding(timeseries_data, freq):
(train, test, target) = timeseries_data
train, valid = train_test_split(train, random_state=42)
data_config = DataConfig(
target=target + ["Occupancy"],
continuous_cols=["Temperature", "Humidity", "Light", "CO2", "HumidityRatio"],
categorical_cols=[],
date_columns=[("date", freq)],
encode_date_columns=True,
)
model_config_params = dict(task="regression")
model_config = CategoryEmbeddingModelConfig(**model_config_params)
trainer_config = TrainerConfig(max_epochs=1, checkpoints=None, early_stopping=None)
optimizer_config = OptimizerConfig()
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
config = tabular_model.config
datamodule = TabularDatamodule(
train=train,
validation=valid,
config=config,
test=test,
)
datamodule.prepare_data()
if freq != "S":
datamodule.setup("fit")
config = datamodule.config
if freq == "H":
assert "_Hour" in datamodule.train.columns
elif freq == "D":
assert "_Dayofyear" in datamodule.train.columns
elif freq == "T":
assert "_Minute" in datamodule.train.columns
elif freq == "S":
try:
datamodule.setup("fit")
assert False
except RuntimeError:
assert True
# from io import BytesIO
# from urllib.request import urlopen
# from zipfile import ZipFile
# import numpy as np
# import pandas as pd
# import pytest
# from sklearn.datasets import fetch_california_housing, fetch_covtype
# def load_timeseries_data():
# url = "https://archive.ics.uci.edu/ml/machine-learning-databases/00357/occupancy_data.zip"
# resp = urlopen(url)
# zipfile = ZipFile(BytesIO(resp.read()))
# train = pd.read_csv(zipfile.open("datatraining.txt"), sep=",")
# val = pd.read_csv(zipfile.open("datatest.txt"), sep=",")
# test = pd.read_csv(zipfile.open("datatest2.txt"), sep=",")
# return (pd.concat([train, val], sort=False), test, ["Occupancy"])
# def load_regression_data():
# dataset = fetch_california_housing(data_home="data", as_frame=True)
# df = dataset.frame.sample(5000)
# df["HouseAgeBin"] = pd.qcut(df["HouseAge"], q=4)
# df["HouseAgeBin"] = "age_" + df.HouseAgeBin.cat.codes.astype(str)
# test_idx = df.sample(int(0.2 * len(df)), random_state=42).index
# test = df[df.index.isin(test_idx)]
# train = df[~df.index.isin(test_idx)]
# return (train, test, dataset.target_names)
# test_dataloader(
# load_regression_data(),
# validation_split=None,
# multi_target=False,
# continuous_cols=[
# "AveRooms",
# "AveBedrms",
# "Population",
# "AveOccup",
# "Latitude",
# "Longitude",
# ],
# categorical_cols=[],
# continuous_feature_transform="yeo-johnson",
# normalize_continuous_features=False,
# target_transform=None,
# embedding_dims=None
# )
# test_date_encoding(load_timeseries_data(), "S")
# ["H","D", "T", "S"],
def main():
# Generate Synthetic Data
data, cat_col_names, num_col_names = data_load()
bsize = 1024
# ##########Define the Configs############
data_config = DataConfig(target=["target"],
continuous_cols=num_col_names,
categorical_cols=cat_col_names)
trainer_config = TrainerConfig(auto_lr_find=True,
batch_size=bsize,
max_epochs=100,
gpus=1)
optimizer_config = OptimizerConfig()
model_config = CategoryEmbeddingModelConfig(task="classification",
layers="1024-512-512",
activation="LeakyReLU",
learning_rate=1e-3)
tabular_mode = TabularModel(data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config)
# Training the Model
# tabular_mode.fit(train=train, validation=val)
# # Evaluating the Model
# # #Loss and Metrics on New Data¶
# result = tabular_mode.evaluate(test)
cv = StratifiedKFold(n_splits=10, shuffle=True)
res_pred = []
for train_idx, test_idx in cv.split(X=data, y=data.target.values):
train, test = data.iloc[train_idx], data.iloc[test_idx]
train, val = train_test_split(train, random_state=42)
tabular_mode = TabularModel(data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config)
weighted_loss = get_class_weighted_cross_entropy(
train["target"].values.ravel(), mu=0.1)
# Training the Model
tabular_mode.fit(train=train,
validation=val,
max_epochs=100,
loss=weighted_loss)
pred_df = tabular_mode.predict(test).loc[:, ["prediction"]]
res_pred.append(pred_df)
# #New Predictions as DataFrame
pred_tot = pd.concat(res_pred).sort_index()
print_metrics(data['target'], pred_tot["prediction"], tag="Holdout")
confusion_matrix(data['target'], pred_tot["prediction"])
# saving model
tabular_mode.save_model("Analysis/basic")
categorical_cols=cat_col_names,
continuous_feature_transform="quantile_normal",
normalize_continuous_features=True)
trainer_config = TrainerConfig(
auto_lr_find=
True, # Runs the LRFinder to automatically derive a learning rate
batch_size=1024,
max_epochs=1000,
gpus=1, #index of the GPU to use. 0, means CPU
)
optimizer_config = OptimizerConfig()
model_config = CategoryEmbeddingModelConfig(
task="classification",
layers="4096-4096-512", # Number of nodes in each layer
activation="LeakyReLU", # Activation between each layers
learning_rate=1e-3,
metrics=["accuracy", "f1"],
metrics_params=[{}, {
"average": "micro"
}])
tabular_model = TabularModel(
data_config=data_config,
model_config=model_config,
optimizer_config=optimizer_config,
trainer_config=trainer_config,
)
# %%
tabular_model.fit(train=train, test=test)
# %%
