2020-06-14 17:03:02 +00:00
|
|
|
import onnx
|
2020-05-19 15:36:04 +00:00
|
|
|
import numpy as np
|
|
|
|
|
import time
|
|
|
|
|
|
2020-07-14 16:17:14 +00:00
|
|
|
init_temperature = 10
|
|
|
|
|
final_temperature_step = 50
|
|
|
|
|
heightened_final_step = 90
|
|
|
|
|
heightened_temp_min = .1
|
2020-06-14 17:03:02 +00:00
|
|
|
|
2020-07-14 16:17:14 +00:00
|
|
|
for step in range(100):
|
|
|
|
|
temp = max(1, 1 + init_temperature * (final_temperature_step - step) / final_temperature_step)
|
|
|
|
|
if temp == 1 and step > final_temperature_step and heightened_final_step and heightened_final_step != 1:
|
|
|
|
|
# Once the temperature passes (1) it enters an inverted curve to match the linear curve from above.
|
|
|
|
|
# without this, the attention specificity "spikes" incredibly fast in the last few iterations.
|
|
|
|
|
h_steps_total = heightened_final_step - final_temperature_step
|
|
|
|
|
h_steps_current = min(step - final_temperature_step, h_steps_total)
|
|
|
|
|
# The "gap" will represent the steps that need to be traveled as a linear function.
|
|
|
|
|
h_gap = 1 / heightened_temp_min
|
|
|
|
|
temp = h_gap * h_steps_current / h_steps_total
|
|
|
|
|
# Invert temperature to represent reality on this side of the curve
|
|
|
|
|
temp = 1 / temp
|
|
|
|
|
print("%i: %f" % (step, temp))
|
