github/concrete
1
1
Fork 0
mirror of https://github.com/zama-ai/concrete.git synced 2026-02-08 19:44:57 -05:00
Code Issues Packages Projects Releases Wiki Activity

fix: select correct parameters for dequantization after evaluation in logistic regression benchmark

Browse Source
This commit is contained in:
Umut
2021-09-30 12:34:18 +03:00
parent a9d44f4758
commit edcbc0cffd
1 changed files with 13 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
benchmarks/logistic_regression.py
View File

@@ -158,14 +158,17 @@ def main():
intermediate = x @ w + b
intermediate_q = x_q.affine(w_q, b_q, intermediate.min(), intermediate.max(), output_bits)
n_y = output_bits
q_y = (2 ** output_bits - 1) / (intermediate.max() - intermediate.min())
zp_y = int(round(intermediate.min() * q_y))
y_parameters = QuantizationParameters(q_y, zp_y, n_y)
sigmoid = QuantizedFunction.plain(
lambda x: 1 / (1 + np.exp(-x)), intermediate_q.parameters, output_bits
)
y_q = sigmoid.apply(intermediate_q)
y_parameters = y_q.parameters
q_x = x_q.parameters.q
q_w = w_q.parameters.q
q_b = b_q.parameters.q
q_intermediate = intermediate_q.parameters.q
zp_x = x_q.parameters.zp
zp_w = w_q.parameters.zp
@@ -175,10 +178,10 @@ def main():
w_q = w_q.values
b_q = b_q.values
c1 = q_y / (q_x * q_w)
c1 = q_intermediate / (q_x * q_w)
c2 = w_q + zp_w
c3 = (q_x * q_w / q_b) * (b_q + zp_b)
c4 = intermediate.min() * q_y
c4 = intermediate.min() * q_intermediate
def f(x):
values = ((c1 * (x + c3)) - c4).round().clip(0, 2 ** output_bits - 1).astype(np.uint)
@@ -229,7 +232,10 @@ def main():
if prediction == y_i:
correct += 1
# Measure: Accuracy (%) = (correct / len(y)) * 100
accuracy = (correct / len(y)) * 100
print(f"Accuracy: {accuracy:.2f}%")
# Measure: Accuracy (%) = accuracy
if __name__ == "__main__":