select_binary, validate_select_binary)
from runtime.tensorflow.explain import explain
from runtime.tensorflow.predict import pred
from runtime.tensorflow.train import train
if __name__ == "__main__":
# Train and explain BoostedTreesClassifier
train(datasource=datasource,
estimator_string="tf.estimator.BoostedTreesClassifier",
select="SELECT * FROM iris.train where class!=2",
validation_select="SELECT * FROM iris.test where class!=2",
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta=label_meta,
model_params={
"n_batches_per_layer": 1,
"n_classes": 2,
"n_trees": 50,
"center_bias": True
},
save="btmodel",
batch_size=100,
epoch=20,
verbose=0)
explain(datasource=datasource,
estimator_string="tf.estimator.BoostedTreesClassifier",
select="SELECT * FROM iris.test where class!=2",
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta = {
"feature_name": "target",
"dtype": "float32",
"delimiter": "",
"shape": [],
"is_sparse": "false" == "true"
}
if __name__ == "__main__":
train(datasource=datasource,
estimator_string="sqlflow_models.DNNRegressor",
select=select,
validation_select=validation_select,
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta=label_meta,
model_params={"hidden_units": [10, 20]},
validation_metrics=["CategoricalAccuracy"],
save="myregmodel_keras",
batch_size=1,
epoch=3,
verbose=0)
pred(datasource=datasource,
estimator_string="sqlflow_models.DNNRegressor",
select=validation_select,
result_table="housing.predict",
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_column_names_map=feature_column_names_map,
train_label_name=label_meta["feature_name"],
result_col_name=label_meta["feature_name"],
from estimator_example import (datasource, feature_column_names,
feature_columns, feature_metas, label_meta)
from runtime.tensorflow.evaluate import evaluate
from runtime.tensorflow.train import train
if __name__ == "__main__":
# Test evaluation on an estimator model
train(datasource=datasource,
estimator_string="tf.estimator.DNNClassifier",
select="SELECT * FROM iris.train where class!=2",
validation_select="SELECT * FROM iris.test where class!=2",
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta=label_meta,
model_params={
"n_classes": 2,
"hidden_units": [128, 32]
},
save="bin_model",
batch_size=10,
epoch=20,
verbose=0)
# FIXME(typhoonzero): need to re-create result table: iris.evaluate_result?
evaluate(
datasource=datasource,
estimator_string="tf.estimator.DNNClassifier",
select="SELECT * FROM iris.test where class!=2",
result_table="",
feature_columns=feature_columns,
feature_column_names=feature_column_names,
"delimiter": "",
"shape": [],
"is_sparse": "false" == "true"
}
if __name__ == "__main__":
# tf.python.training.basic_session_run_hooks.LoggingTensorHook
# = runtime.tensorflow.train.PrintTensorsHook
train(datasource=datasource,
estimator_string="DNNClassifier",
select=select,
validation_select=validate_select,
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta=label_meta,
model_params={
"n_classes": 3,
"hidden_units": [10, 20]
},
save="mymodel",
batch_size=1,
epoch=3,
verbose=0)
train(datasource=datasource,
estimator_string="DNNClassifier",
select=select_binary,
validation_select=validate_select_binary,
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta=label_meta,
feature_columns,
feature_metas, label_meta,
select, validate_select)
from runtime.tensorflow.predict import pred
from runtime.tensorflow.train import train
if __name__ == "__main__":
train(datasource=datasource,
estimator_string="sqlflow_models.DNNClassifier",
select=select,
validation_select=validate_select,
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_metas=feature_metas,
label_meta=label_meta,
model_params={
"n_classes": 3,
"hidden_units": [10, 20]
},
validation_metrics=["CategoricalAccuracy"],
save="mymodel_keras",
batch_size=1,
epoch=3,
verbose=0)
pred(datasource=datasource,
estimator_string="sqlflow_models.DNNClassifier",
select=select,
result_table="iris.predict",
feature_columns=feature_columns,
feature_column_names=feature_column_names,
feature_column_names_map=feature_column_names_map,
def do_train(params):
train.train(params["datasource"], params["estimator_string"],
params["select"], params["validation_select"],
params["feature_columns"], params["feature_column_names"],
**params)