68 lines
2.3 KiB
Python
68 lines
2.3 KiB
Python
# Sebastian Raschka 2014-2020
|
|
# mlxtend Machine Learning Library Extensions
|
|
#
|
|
# Base Clusteer (Clutering Parent Class)
|
|
# Author: Sebastian Raschka <sebastianraschka.com>
|
|
#
|
|
# License: BSD 3 clause
|
|
|
|
import numpy as np
|
|
|
|
|
|
class _MultiLayer(object):
|
|
|
|
def __init__(self):
|
|
pass
|
|
|
|
def _layermapping(self, n_features, n_classes, hidden_layers):
|
|
"""Creates a dictionaries of layer dimensions for weights and biases.
|
|
|
|
For example, given
|
|
`n_features=10`, `n_classes=10`, and `hidden_layers=[8, 7, 6]`:
|
|
|
|
biases =
|
|
{1: [[8], 'n_hidden_1'],
|
|
2: [[7], 'n_hidden_2'],
|
|
3: [[6], 'n_hidden_3'],
|
|
'out': [[10], 'n_classes']
|
|
}
|
|
|
|
weights =
|
|
{1: [[10, 8], 'n_features, n_hidden_1'],
|
|
2: [[8, 7], 'n_hidden_1, n_hidden_2'],
|
|
3: [[7, 6], 'n_hidden_2, n_hidden_3'],
|
|
'out': [[6, 10], 'n_hidden_3, n_classes']
|
|
}
|
|
|
|
"""
|
|
weights = {'1': [[n_features, hidden_layers[0]],
|
|
'n_features, n_hidden_1'],
|
|
'out': [[hidden_layers[-1], n_classes],
|
|
'n_hidden_%d, n_classes' % len(hidden_layers)]}
|
|
biases = {'1': [[hidden_layers[0]], 'n_hidden_1'],
|
|
'out': [[n_classes], 'n_classes']}
|
|
|
|
if len(hidden_layers) > 1:
|
|
for i, h in enumerate(hidden_layers[1:]):
|
|
layer = i + 2
|
|
weights[str(layer)] = [[weights[str(layer - 1)][0][1], h],
|
|
'n_hidden_%d, n_hidden_%d'
|
|
% (layer - 1, layer)]
|
|
biases[str(layer)] = [[h], 'n_hidden_%d' % layer]
|
|
return weights, biases
|
|
|
|
def _init_params_from_layermapping(self, weight_maps, bias_maps,
|
|
random_seed=None):
|
|
rgen = np.random.RandomState(random_seed)
|
|
weights, biases = {}, {}
|
|
|
|
rgen = np.random.RandomState(random_seed)
|
|
weights, biases = {}, {}
|
|
|
|
for kw, kb in zip(sorted(weight_maps), sorted(bias_maps)):
|
|
weights[kw] = rgen.normal(loc=0.0,
|
|
scale=0.01,
|
|
size=weight_maps[kw][0])
|
|
biases[kb] = np.zeros(shape=bias_maps[kb][0])
|
|
return weights, biases
|