def test_lift3():
# https://www.kaggle.com/vsmolyakov/svm-classifier
# with some mods --> remove deprecated stale/code from sklearn
# in source (from older version, incompatible)
src = """
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import csv as csv
from sklearn.svm import SVC
from sklearn.model_selection import train_test_split, GridSearchCV
from sklearn.preprocessing import StandardScaler
# subbed out data here...
X_train = np.random.random((100, 100))
y_train = np.random.random(100) > 0.5
svm_parameters = [{'kernel': ['rbf'], 'C': [1,10,100,1000]}]
clf = GridSearchCV(SVC(), svm_parameters, cv=3, verbose=2)
clf.fit(X_train, y_train)
clf.best_params_
C_opt = 10
clf = SVC(C=C_opt, kernel='rbf')
clf.fit(X_train, y_train)
clf.n_support_
X_test_data = np.random.random((100, 100))
y_pred = clf.predict(X_test_data)
"""
result = PipelineLifter(src)
assert not result.failed
assert len(result.pipelines) == 2
from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.svm import SVC
clf = SVC(C=10, kernel="rbf")
expected1 = Pipeline([("clf", clf)])
svm_parameters = [{'kernel': ['rbf'], 'C': [1, 10, 100, 1000]}]
clf = GridSearchCV(SVC(), svm_parameters, cv=3, verbose=2)
expected2 = Pipeline([("clf", clf)])
expected = set([pt.md5(expected1), pt.md5(expected2)])
got = set([pt.md5(p) for p in result.pipelines])
assert got == expected
def test_lift2():
# same pipeline but now not using the Pipeline construct in the
# source
src = """
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
logit = LogisticRegression(
C=1,
solver='lbfgs',
max_iter=500,
random_state=17,
n_jobs=1,
multi_class='multinomial')
import numpy as np
X = np.random.random((10, 10))
y = np.random.random(10)
y = y > 0.5
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
logit.fit(X_scaled, y)
logit.predict(X_scaled)
"""
result = PipelineLifter(src)
assert not result.failed
assert len(result.pipelines) == 1
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
logit = LogisticRegression(C=1,
solver='lbfgs',
max_iter=500,
random_state=17,
n_jobs=1,
multi_class='multinomial')
expected = Pipeline([('scaler', StandardScaler()), ('logit', logit)])
assert pt.md5(expected) == pt.md5(result.pipelines[0])
def get_repair_hashes(repairer, seed, num_passes=2, num_pipelines=3):
passes = [[]] * num_passes
for i in range(0, num_passes):
print("Pass: {}".format(i))
utils.set_seed(seed)
num_remaining = num_pipelines
pbar = tqdm.tqdm(total=num_pipelines)
for p in data.pipelines:
if num_remaining <= 0:
break
repaired = repairer.repair(p, data.X, data.y, bound_num_repairs=1)
orig_md5 = pt.md5(p)
if repaired is None:
continue
repaired_md5 = pt.md5(repaired)
if orig_md5 == repaired_md5:
continue
passes[i].append(repaired_md5)
num_remaining -= 1
pbar.update(1)
pbar.close()
return passes
def test_lift1():
src = """
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
logit = LogisticRegression(
C=1,
solver='lbfgs',
max_iter=500,
random_state=17,
n_jobs=1,
multi_class='multinomial')
p = Pipeline([('scaler', StandardScaler()), ('logit', logit)])
import numpy as np
X = np.random.random((10, 10))
y = np.random.random(10)
y = y > 0.5
p.fit(X, y)
p.predict(X)
"""
result = PipelineLifter(src)
assert not result.failed
assert len(result.pipelines) == 1
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
logit = LogisticRegression(C=1,
solver='lbfgs',
max_iter=500,
random_state=17,
n_jobs=1,
multi_class='multinomial')
expected = Pipeline([('scaler', StandardScaler()), ('logit', logit)])
assert pt.md5(expected) == pt.md5(result.pipelines[0])
def extract_pipelines(self):
assert self.graph is not None
seed_node_ids = find_pipeline_seeds(self.graph)
slices = get_graph_slices(self.graph, seed_node_ids)
slices = remove_subgraphs(slices)
path_lengths = nx.all_pairs_shortest_path_length(
self.graph.reverse(copy=True))
path_lengths = dict(path_lengths)
pipelines = []
hashes = set([])
for _slice in slices:
annotated_slice = execute_graph_line_by_line(_slice)
components = extract_pipeline_components(annotated_slice,
path_lengths)
pipeline = build_pipeline(components)
h = pt.md5(pipeline)
if pipeline is not None and h not in hashes:
pipelines.append(pipeline)
hashes.add(h)
return pipelines