networkprocossor. Get better prediction by taking lowest points in graphic

[dieuska]

when there are noflat parts in the diff graphic (stable geometries), maybe good to take the biggest breakpoints?

[dieuska]

import numpy as np
import matplotlib.pyplot as plt
from scipy.signal import argrelextrema

# 1. Generate example data
# We use a function that is always increasing (no 0 slope)
x = np.linspace(0, 10, 500)
y = np.sin(x) + 1.2 * x  

# 2. Calculate the slope (derivative)
dy = np.gradient(y, x)

# 3. Take the absolute value of the slope
# This represents "steepness"
abs_dy = np.abs(dy)

# 4. Find the local minima of the absolute slope
# 'order' defines how many neighboring points to compare with
# Increase 'order' to ignore small fluctuations/noise
flattest_indices = argrelextrema(abs_dy, np.less, order=20)[0]

# --- Visualization ---
plt.figure(figsize=(10, 6))
plt.plot(x, y, label='Data Curve', color='blue')

# Mark the flattest points
plt.scatter(x[flattest_indices], y[flattest_indices], 
            color='red', zorder=5, label='Flattest Points')

plt.title("Detecting Points with Minimum Steepness")
plt.xlabel("X")
plt.ylabel("Y")
plt.legend()
plt.grid(True, alpha=0.3)
plt.show()

[dieuska]

without scipy

import numpy as np
import matplotlib.pyplot as plt

# 1. Generate data (an increasing curve with a "relatively" flat part)
x = np.linspace(0, 10, 500)
y = np.sin(x) + 1.1 * x 

# 2. Calculate the absolute slope
dy = np.gradient(y, x)
abs_dy = np.abs(dy)

# 3. Find local minima without SciPy
# We check if a value is smaller than the one before it AND the one after it
# This creates a boolean mask
is_local_min = (abs_dy[1:-1] < abs_dy[:-2]) & (abs_dy[1:-1] < abs_dy[2:])

# Convert boolean mask to indices (and shift by 1 because we sliced 1:-1)
flattest_indices = np.where(is_local_min)[0] + 1

# --- Optional: Sensitivity Filter ---
# If your data is noisy, 'is_local_min' will find too many points.
# You can filter by a threshold (e.g., only keep the top 3 flattest points)
# sorted_indices = flattest_indices[np.argsort(abs_dy[flattest_indices])][:3]

# 4. Visualization
plt.figure(figsize=(10, 5))
plt.plot(x, y, label='Data')
plt.scatter(x[flattest_indices], y[flattest_indices], color='red', label='Local Flattest Points')
plt.title("Flat Point Detection (No SciPy)")
plt.legend()
plt.show()

[dieuska]

see function "determine_stability2()"