def test_multi_exp_single_flows_single_outputload(self):
flow2 = d6tflow.WorkflowMulti({'experiment1': {'do_preprocess': False}, 'experiment2': {'do_preprocess': True}},
task=TaskTrain)
flow2.run(flow = "experiment1")
out = flow2.outputLoad(TaskTrain)
type(out).__name__ == "LogisticRegression"
def test_output_load_comparing_with_task_run(self):
params2 = {'param1': 2}
params = {'param1': 1}
flow2 = d6tflow.WorkflowMulti({1: params, 2: params2}, task=Task1)
flow2.run()
assert flow2.outputLoad(flow = 1) == Task1(**params).outputLoad()
assert flow2.outputLoad(flow = 2) == Task1(**params2).outputLoad()
def test_multi_exp_all_flows_outputloadall(self):
flow2 = d6tflow.WorkflowMulti({'experiment1': {'do_preprocess': False}, 'experiment2': {'do_preprocess': True}},
task=TaskTrain)
flow2.run()
data = flow2.outputLoadAll()
type(data['experiment1']['TaskTrain']).__name__ == "LogisticRegression"
type(data['experiment2']['TaskTrain']).__name__ == "LogisticRegression"
def test_output_load_all_comparing_with_task_run(self):
params2 = {'param1': 2}
params = {'param1': 1}
flow2 = d6tflow.WorkflowMulti({1: params, 2: params2}, task=Task1)
flow2.run()
outputs = flow2.outputLoadAll()
assert outputs[1]['Task1'] == Task1(**params).outputLoad()
assert outputs[2]['Task1'] == Task1(**params2).outputLoad()
def test_multi_get_task(self):
params2 = {'param1': 2}
params = {'param1': 1}
flow2 = d6tflow.WorkflowMulti(params={
1: params,
2: params2
},
task=Task1)
assert flow2.get_task(flow=1).param_kwargs['param1'] == 1
assert flow2.get_task(flow=2).param_kwargs['param1'] == 2
def test_no_default_expect_error(self):
with pytest.raises(RuntimeError):
flow2 = d6tflow.WorkflowMulti(
params={
'experiment1': {
'do_preprocess': False
},
'experiment2': {
'do_preprocess': True
}
})
flow2.run()
def test_default_task(self):
flow2 = d6tflow.WorkflowMulti(
params={
'experiment1': {
'do_preprocess': False
},
'experiment2': {
'do_preprocess': True
}
})
flow2.set_default(TaskTrain)
out_class = flow2.get_task()
type(out_class).__name__ == "TaskTrain"
def test_preview_single_flow(self):
flow2 = d6tflow.WorkflowMulti({'experiment1': {'do_preprocess': False}, 'experiment2': {'do_preprocess': True}},
task=TaskTrain)
import io
from contextlib import redirect_stdout
with io.StringIO() as buf, redirect_stdout(buf):
flow2.preview(TaskTrain, flow = 'experiment1')
output = buf.getvalue()
assert output.count('PENDING')==0
assert output.count('COMPLETE')==3
assert output.count("'do_preprocess': 'False'")==1
def test_path():
class Task1(d6tflow.tasks.TaskPickle):
def run(self):
self.save({1: 1})
class Task2(d6tflow.tasks.TaskPickle):
def run(self):
self.save({1: 1})
path = 'data/data2/'
assert 'data2' in str(Task1(path=path).output().path)
flow = d6tflow.Workflow(Task1, path=path)
assert 'data2' in str(flow.get_task().output().path)
flow2 = d6tflow.WorkflowMulti(Task2, params={0: {}}, path=path)
assert 'data2' in str(flow2.get_task()[0].output().path)
model = sklearn.ensemble.GradientBoostingRegressor()
else:
raise ValueError('invalid model selection')
model.fit(df_train.drop('y', 1), df_train['y'])
self.save(model)
self.saveMeta(
{'score': model.score(df_train.drop('y', 1), df_train['y'])})
# goal: compare performance of two models
params_model1 = {'do_preprocess': True, 'model': 'ols'}
params_model2 = {'do_preprocess': False, 'model': 'gbm'}
# run workflow
flow = d6tflow.WorkflowMulti(ModelTrain, {
'ols': params_model1,
'gbm': params_model2
})
flow.reset_upstream(confirm=False) # force re-run
print(flow.preview('ols'))
flow.run()
'''
Scheduled 3 tasks of which:
* 3 ran successfully:
- 1 GetData()
- 1 ModelData(do_preprocess=True)
- 1 ModelTrain(do_preprocess=True, model=ols)
# To run 2nd model, don't need to re-run all tasks, only the ones that changed
Scheduled 3 tasks of which:
* 1 complete ones were encountered:
elif self.model == 'gbm':
model = sklearn.ensemble.GradientBoostingRegressor()
# fit and save model with training score
model.fit(df_trainX, df_trainY)
self.save(model) # persist/cache model
self.saveMeta({'score': model.score(df_trainX,
df_trainY)}) # save model score
# goal: compare performance of two models
# define workflow manager
flow = d6tflow.WorkflowMulti(ModelTrain, {
'model1': {
'model': 'ols'
},
'model2': {
'model': 'gbm'
}
})
flow.reset_upstream(confirm=False) # DEMO ONLY: force re-run
flow.run() # execute model training including all dependencies
'''
===== Execution Summary =====
Scheduled 2 tasks of which:
* 2 ran successfully:
- 1 GetData()
- 1 ModelTrain(model=ols)
This progress looks :) because there were no failed tasks or missing dependencies
'''
scores = flow.outputLoadMeta() # load model scores
def test_no_exp_params_passed(self):
with pytest.raises(Exception) as e:
flow2 = d6tflow.WorkflowMulti()
assert str(e.value) == "Experments not defined"