文件列表:

LICENSE (1067, 2019-02-24)
cnn/ (0, 2019-02-24)
cnn/activations.py (685, 2019-02-24)
cnn/convolution.py (2146, 2019-02-24)
cnn/derivatives.py (695, 2019-02-24)
cnn/derivatives_losses.py (644, 2019-02-24)
cnn/layer_convolutional.py (3656, 2019-02-24)
cnn/layer_dense.py (3754, 2019-02-24)
cnn/losses.py (607, 2019-02-24)
cnn/nn.py (16839, 2019-02-24)
cnn/parameters_init.py (2087, 2019-02-24)
cnn/pooling.py (1844, 2019-02-24)
cnn/series_to_matrix.py (4029, 2019-02-24)
cnn/utils.py (250, 2019-02-24)
test/ (0, 2019-02-24)
test/experiments/ (0, 2019-02-24)
test/experiments/data/ (0, 2019-02-24)
test/experiments/data/TOPIX_index.csv (21917, 2019-02-24)
test/experiments/data/naive_dataset.csv (48200, 2019-02-24)
test/experiments/data/power_consumption.csv (158861, 2019-02-24)
test/experiments/data/sin.csv (165130, 2019-02-24)
test/experiments/data/space_shuttle_marotta_valve.csv (85000, 2019-02-24)
test/experiments/data/test_lag.csv (2251, 2019-02-24)
test/experiments/data_utils/ (0, 2019-02-24)
test/experiments/data_utils/utils_topix.py (4917, 2019-02-24)
test/experiments/prediction_naive_dataset_test_gaussian.py (4719, 2019-02-24)
test/experiments/prediction_power_consumption.py (5309, 2019-02-24)
test/experiments/prediction_sin_test_gaussian.py (4482, 2019-02-24)
test/experiments/prediction_space_shuttle_gaussian.py (4811, 2019-02-24)
test/experiments/prediction_space_shuttle_randomness.py (5268, 2019-02-24)
test/experiments/prediction_test_lag_gaussian.py (4346, 2019-02-24)
test/experiments/prediction_topix_gaussian.py (4723, 2019-02-24)
test/experiments/prediction_topix_test_autocorrelation.py (5365, 2019-02-24)
test/experiments/prediction_topix_test_randomness.py (5318, 2019-02-24)
test/experiments/results/ (0, 2019-02-24)
test/experiments/results/SPACE_gaussian_anomaly_threshold_mu-plus-2sigma_conv_lrelu_25-2-2_conv_lrelu_55-2-2_dense_tanh_75_dense_tanh_1_params_-1e-2_1e1_lrate_1e-2_loss_L2_train_07_valid_05.pdf (21750, 2019-02-24)
test/experiments/results/SPACE_rand_anomaly_alpha_2e-2_chunk_size_15_ninput_10_conv_lrelu_25-2-2_conv_lrelu_15-2-2_dense_tanh_75_dense_tanh_1_params_-1e-2_1e1_lrate_1e-2_loss_L2_train_05_valid_05.pdf (17078, 2019-02-24)
test/experiments/results/TOPIX_corr_anomaly_toler_2e-1_chunk_size_25_ninput_5_conv_lrelu_25-2-2_conv_lrelu_55-2-2_dense_tanh_75_dense_tanh_1_params_-1e-2_1e1_lrate_1e-2_loss_L2_train_07_valid_075.pdf (14909, 2019-02-24)
... ...

# Convolutional Neural Networks for Time Series Analysis ## Package to instantiate 1-D CNNs and train them on signals datasets ## Things implemented so far: ## - customizable deep net (specify with one line of code the net, from the topology to the activations); - all from backpropagation to check gradient routine depends only on numpy; - easy to extend activations, derivatives, parameters initialization etc. TODO: - refactor code and package's name to handle 2D/3D images; - topologies that mix layers of convolutionals and dense type: so far it is possible to specify n>0 convolutional layers followed by m>0 dense layers. ## Create a double-layer CNN followed by a final double-dense layer ```python import nn as nn # nn.py from this package import numpy as np # define networks layers net_blocks = {'n_inputs': 50, 'layers': [ {'type': 'conv', 'activation': 'leaky_relu', 'shape': (15, 2), 'stride': 3}, {'type': 'conv', 'activation': 'leaky_relu', 'shape': (30, 2), 'stride': 3}, {'type': 'dense', 'activation': 'relu', 'shape': (None, 75)}, {'type': 'dense', 'activation': 'relu', 'shape': (None, 1)} ] } # create the net net = nn.NN(net_blocks) # initialize the parameters net.init_parameters(['uniform', -.1e-1, 1e-1]) ``` Now you can check one of the test in the /test/experiments forlder to see how to run this model against a dataset.

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：