chatgpt-4o-latest-20250326· OpenAI
LMArena Human Preference 140K
impl the idea python normales = np.tile(invalid, (N, 2)) normales[0, 0:1] = invalid #…
- User
impl the idea python
normales = np.tile(invalid, (N, 2)) normales[0, 0:1] = invalid # (a,b) normales[0, 2:3] = segment_angles_list[0].to_numpy() normales[1, 0:1] = segment_angles_list[0].to_numpy() normales[1, 2:3] = segment_angles_list[1].to_numpy() ... normales[last, 0:1] = segment_angles_list[last].to_numpy() normales[last, 2:3] = invalid # (a,b) return normales - Assistant
Thanks for the clarification! You're trying to construct a
normalesarray that contains overlapping pairs of angles, such that each row contains:[prev_angle.a, prev_angle.b, current_angle.a, current_angle.b]And you wrap it with invalids at the beginning and end.
Let’s implement this idea in Python properly — taking into account that:
segment_angles_listis likely a list ofAngleobjects each having.to_numpy()→[sin, cos]- You want a resulting NumPy array shaped
(len(segment_angles_list) + 1, 4)(for [a, b, a, b] format) invalidshould be a 2-element NumPy array, e.g.np.array([np.nan, np.nan])or similar
✅ Final Working Implementation
import numpy as np def build_normales(segment_angles_list, invalid): num_segments = len(segment_angles_list) N = num_segments + 1 # We need one more row than the number of segments # Each row will have 4 columns: [prev.sin, prev.cos, curr.sin, curr.cos] normales = np.tile(invalid, (N, 2)) # Shape: (N, 4) if invalid is shape (2,) # First row: [invalid, angle_0] normales[0, 0:2] = invalid normales[0, 2:4] = segment_angles_list[0].to_numpy() # Intermediate rows for i in range(1, num_segments): normales[i, 0:2] = segment_angles_list[i - 1].to_numpy() normales[i, 2:4] = segment_angles_list[i].to_numpy() # Last row: [angle_last, invalid] normales[N - 1, 0:2] = segment_angles_list[-1].to_numpy() normales[N - 1, 2:4] = invalid return normales
Example usage
from dataclasses import dataclass import math import numpy as np @dataclass class Angle: sin: float cos: float def to_numpy(self): return np.array([self.sin, self.cos], dtype=np.float32) # Test input angles = [ Angle(math.sin(0), math.cos(0)), Angle(math.sin(0.5), math.cos(0.5)), Angle(math.sin(1.0), math.cos(1.0)), Angle(math.sin(1.5), math.cos(1.5)) ] invalid = np.array([np.nan, np.nan], dtype=np.float32) normales = build_normales(angles, invalid) print(normales)
Output
[[ nan nan 0. 1. ] [ 0. 1. 0.47942555 0.87758255] [ 0.47942555 0.87758255 0.84147096 0.5403023 ] [ 0.84147096 0.5403023 0.99749494 0.0707372 ] [ 0.99749494 0.0707372 nan nan]]This gives you exactly the format you're looking for 😊
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.