"""
Copyright (C) 2018 Patrick Schwab, ETH Zurich
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
documentation files (the "Software"), to deal in the Software without restriction, including without limitation
the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software,
and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions
of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED
TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF
CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.
"""
import tensorflow as tf
# SOURCE: https://github.com/clinicalml/cfrnet, MIT-License
def pdist2(X, Y):
""" Computes the squared Euclidean distance between all pairs x in X, y in Y """
C = -2*tf.matmul(X, tf.transpose(Y))
nx = tf.reduce_sum(tf.square(X), 1, keep_dims=True)
ny = tf.reduce_sum(tf.square(Y), 1, keep_dims=True)
D = (C + tf.transpose(ny)) + nx
return tf.sqrt(D + 1e-8)
# SOURCE: https://github.com/clinicalml/cfrnet, MIT-License
def cf_nn(x, t):
It = tf.where(tf.equal(t, 1))[:, 0]
Ic = tf.where(tf.equal(t, 0))[:, 0]
x_c = tf.gather(x, Ic)
x_t = tf.gather(x, It)
D = pdist2(x_c, x_t)
nn_t = tf.gather(Ic, tf.argmin(D, 0))
nn_c = tf.gather(It, tf.argmin(D, 1))
return tf.stop_gradient(nn_t), tf.stop_gradient(nn_c)
# SOURCE: https://github.com/clinicalml/cfrnet, MIT-License
def pehe_nn(yf_p, ycf_p, y, x, t, nn_t=None, nn_c=None):
if nn_t is None or nn_c is None:
nn_t, nn_c = cf_nn(x, t)
It = tf.where(tf.equal(t, 1))[:, 0]
Ic = tf.where(tf.equal(t, 0))[:, 0]
ycf_t = 1.0*tf.gather(y, nn_t)
eff_nn_t = ycf_t - 1.0*tf.gather(y, It)
eff_pred_t = tf.gather(ycf_p, It) - tf.gather(yf_p, It)
eff_pred = eff_pred_t
eff_nn = eff_nn_t
pehe_nn = tf.sqrt(tf.reduce_mean(tf.square(eff_pred - eff_nn)))
return pehe_nn
def pehe_loss(y_true, y_pred, t, x, num_treatments):
total, num_elements = 0, 0.
for i in range(num_treatments):
for j in range(num_treatments):
if j >= i:
continue
t1_indices = tf.where(tf.equal(t, i))[:, 0]
t2_indices = tf.where(tf.equal(t, j))[:, 0]
these_x = tf.concat([tf.gather(x, t1_indices), tf.gather(x, t2_indices)], axis=0)
y_pred_these_treatments = tf.concat([tf.gather(y_pred, t1_indices),
tf.gather(y_pred, t2_indices)], axis=0)
y_true_these_treatments = tf.concat([tf.gather(y_true, t1_indices),
tf.gather(y_true, t2_indices)], axis=0)
these_treatments = tf.concat([tf.ones((tf.shape(t1_indices)[0],), dtype="int32") * i,
tf.ones((tf.shape(t2_indices)[0],), dtype="int32") * j],
axis=0)
these_y_pred_f = tf.gather(y_pred_these_treatments,
tf.concat([tf.range(tf.shape(y_pred_these_treatments)[0]),
these_treatments], axis=-1))
these_y_true_f = y_true_these_treatments
inverse_treatments = tf.concat([tf.ones((tf.shape(t1_indices)[0],), dtype="int32") * j,
tf.ones((tf.shape(t2_indices)[0],), dtype="int32") * i],
axis=0)
these_y_pred_cf = tf.gather(y_pred_these_treatments,
tf.concat([tf.range(tf.shape(y_pred_these_treatments)[0]),
inverse_treatments], axis=-1))
these_treatments = tf.concat([tf.zeros((tf.shape(t1_indices)[0],), dtype="int32"),
tf.ones((tf.shape(t2_indices)[0],), dtype="int32")],
axis=0)
total += pehe_nn(these_y_pred_f, these_y_pred_cf, these_y_true_f, these_x, these_treatments)
num_elements += 1.
return total / num_elements
