def build_proxy(self, *, source, trained):
self.data[("value_predicted", "classes")] = ProxyContainer(
lambda : trained.data[("value_statistics", "classes")],
Header,
)
self.data[("value_predicted", "probability")] = ProxyContainer(
lambda: trained.data[("value_statistics", "probability")].reshape(1,-1) * np.ones(source.data["representation"].shape[0], dtype=np.float).reshape(-1,1),
Table,
)
def build_proxy(self, *, source, trained):
rfc = trained.rfc.fit(
trained.source.data["representation"],
trained.source.data["value"],
)
self.data[("value_predicted", "classes")] = ProxyContainer(
lambda: rfc.classes_,
Header,
)
self.data[("value_predicted", "probability")] = ProxyContainer(
lambda: rfc.predict_proba(source.data["representation"]),
Table,
)
def build_proxy(self, *, source, trained):
self.data["value_predicted"] = ProxyContainer(
lambda : np.sum(
source.data["representation"] * trained.data["w"],
axis=1,
) + trained.data["b"],
Table,
)
def build_proxy(self, *, source, trained):
etr = trained.etr.fit(
trained.source.data["representation"],
trained.source.data["value"],
)
self.data["value_predicted"] = ProxyContainer(
lambda: etr.predict(source.data["representation"]),
Table,
)
def build_proxy(self, *, source, trained):
self.data["value_predicted"] = ProxyContainer(
lambda : np.array(
[np.mean(trained.data["value_original"][_idx]) \
for _idx in np.argsort(
-source.data["representation"],
axis=1,
)[:,:trained.data["k"]]]
),
Table,
)
def build_proxy(self, *, source, trained):
self.data[("value_predicted", "classes")] = ProxyContainer(
lambda: np.array([0, 1], dtype=np.int),
Header,
)
def f():
x = source.data["representation"]
w = trained.data["w"]
b = trained.data["b"]
# predict value == 1
result = np.zeros((x.shape[0], 2), dtype=np.float)
result[:, 1] = expit(np.sum(x * w, axis=1) + b)
result[:, 0] = 1 - result[:, 1]
return result
self.data[("value_predicted", "probability")] = ProxyContainer(
f,
Table,
)
def build_proxy(self, *, source, trained):
self.data[("value_predicted", "classes")] = ProxyContainer(
lambda: np.unique(trained.data["value_original"]),
Header,
)
def f():
k = trained.data["k"]
idx = np.argsort(-source.data["representation"], axis=1)[:, :k]
classes = np.unique(trained.data["value_original"])
result = np.zeros((len(idx), len(classes)), dtype=np.float)
for i, _idx in enumerate(idx):
for j, c in enumerate(classes):
result[i, j] = (trained.data["value_original"][_idx]
== c).sum() / k
result[:, -1] = 1 - result[:, :-1].sum(axis=1)
return result
self.data[("value_predicted", "probability")] = ProxyContainer(
f,
Table,
)
def build_proxy(self, *, source, trained):
self.data["value_predicted"] = ProxyContainer(
lambda: trained.data["value_constant"] * np.ones(
source.data["representation"].shape[0], dtype=np.float32),
Table,
)