maximuspowers/muat-fourier-3-classifier

Pattern Classifier

This model was trained to classify which patterns a subject model was trained on, based on neuron activation signatures.

Dataset

• Training Dataset: maximuspowers/muat-fourier-3
• Input Mode: signature
• Number of Patterns: 14

Patterns

The model predicts which of the following 14 patterns the subject model was trained on:

1. palindrome
2. sorted_ascending
3. sorted_descending
4. alternating
5. contains_abc
6. starts_with
7. ends_with
8. no_repeats
9. has_majority
10. increasing_pairs
11. decreasing_pairs
12. vowel_consonant
13. first_last_match
14. mountain_pattern

Model Architecture

• Signature Encoder: [512, 256, 256, 128]
• Activation: relu
• Dropout: 0.2
• Batch Normalization: True

Training Configuration

• Optimizer: adam
• Learning Rate: 0.001
• Batch Size: 16
• Loss Function: BCE with Logits (with pos_weight for training, unweighted for validation)

Test Set Performance

• F1 Macro: 0.2985
• F1 Micro: 0.2756
• Hamming Accuracy: 0.7292
• Exact Match Accuracy: 0.0250
• BCE Loss: 0.4709

Per-Pattern Performance (Test Set)

Pattern	Precision	Recall	F1 Score
palindrome	14.1%	90.0%	24.4%
sorted_ascending	49.1%	62.1%	54.8%
sorted_descending	12.9%	89.7%	22.5%
alternating	18.6%	69.9%	29.4%
contains_abc	26.4%	73.7%	38.8%
starts_with	9.6%	84.7%	17.3%
ends_with	15.1%	82.2%	25.5%
no_repeats	12.7%	59.3%	21.0%
has_majority	56.2%	34.6%	42.9%
increasing_pairs	27.6%	66.7%	39.0%
decreasing_pairs	16.5%	80.8%	27.4%
vowel_consonant	12.1%	50.0%	19.5%
first_last_match	21.1%	78.1%	33.2%
mountain_pattern	13.8%	54.3%	22.0%

Usage

import torch
from huggingface_hub import hf_hub_download

# Download the model
checkpoint_path = hf_hub_download(repo_id='maximuspowers/muat-fourier-3-classifier', filename='best_model.pt')
checkpoint = torch.load(checkpoint_path)