nmelone/Weather-Project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Reconfigured folder structure

Browse Source 
Added Evaluation, Preprocessing, Training, Transformation folders.
Preprocessing is just a rework of the folder for the new structure of the old preprocessing folder.

Training and Transformation are the old project file broken up into two parts and restructured.

Evaluation is for evaluating the predictive power of the model.
This commit is contained in:
nmelone
2018-09-24 20:48:07 -05:00
parent 953507a22a
commit 7b30c71b53
27 changed files with 4547 additions and 4751 deletions
Download Patch File Download Diff File Expand all files Collapse all files

240
Project Final/Evaluation/Evalutation.ipynb Normal file
View File

@@ -0,0 +1,240 @@
{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using TensorFlow backend.\n"
]
}
],
"source": [
"import os\n",
"import numpy as np\n",
"from six.moves import cPickle\n",
"import matplotlib\n",
"matplotlib.use('Agg')\n",
"import matplotlib.pyplot as plt\n",
"import matplotlib.gridspec as gridspec\n",
"%matplotlib inline\n",
"from keras import backend as K\n",
"from keras.models import Model, model_from_json\n",
"from keras.layers import Input, Dense, Flatten\n",
"\n",
"from prednet import PredNet\n",
"from data_utils import SequenceGenerator\n",
"\n",
"from tqdm import tqdm"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"n_plot = 40\n",
"batch_size = 10\n",
"nt = 24"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"WEIGHTS_DIR = '../Training/weights/'\n",
"DATA_DIR = '../data/'\n",
"RESULTS_SAVE_DIR = './weather_results/'"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"weights_file = os.path.join(WEIGHTS_DIR, 'prednet_weather_weights.hdf5')\n",
"json_file = os.path.join(WEIGHTS_DIR, 'prednet_weather_model.json')\n",
"test_file = os.path.join(DATA_DIR, 'x_test.hkl')\n",
"test_sources = os.path.join(DATA_DIR, 'sources_test.hkl')"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"# Load trained model\n",
"f = open(json_file, 'r')\n",
"json_string = f.read()\n",
"f.close()\n",
"train_model = model_from_json(json_string, custom_objects = {'PredNet': PredNet})\n",
"train_model.load_weights(weights_file)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [],
"source": [
"# Create testing model (to output predictions)\n",
"layer_config = train_model.layers[1].get_config()\n",
"layer_config['output_mode'] = 'prediction'\n",
"data_format = layer_config['data_format'] if 'data_format' in layer_config else layer_config['dim_ordering']\n",
"test_prednet = PredNet(weights=train_model.layers[1].get_weights(), **layer_config)\n",
"input_shape = list(train_model.layers[0].batch_input_shape[1:])\n",
"input_shape[0] = nt\n",
"inputs = Input(shape=tuple(input_shape))\n",
"predictions = test_prednet(inputs)\n",
"test_model = Model(inputs=inputs, outputs=predictions)"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [],
"source": [
"test_generator = SequenceGenerator(test_file, test_sources, nt, sequence_start_mode='unique', data_format=data_format)\n",
"X_test = test_generator.create_all()\n",
"X_hat = test_model.predict(X_test, batch_size)\n",
"if data_format == 'channels_first':\n",
" X_test = np.transpose(X_test, (0, 1, 3, 4, 2))\n",
" X_hat = np.transpose(X_hat, (0, 1, 3, 4, 2))"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [],
"source": [
"# Compare MSE of PredNet predictions vs. using last frame. Write results to prediction_scores.txt\n",
"mse_model = np.nanmean( (X_test[:, 1:] - X_hat[:, 1:])**2 ) # look at all timesteps except the first\n",
"mse_prev = np.nanmean( (X_test[:, :-1] - X_test[:, 1:])**2 )\n",
"if not os.path.exists(RESULTS_SAVE_DIR): os.mkdir(RESULTS_SAVE_DIR)\n",
"f = open(RESULTS_SAVE_DIR + 'prediction_scores.txt', 'w')\n",
"f.write(\"Model MSE: %f\\n\" % mse_model)\n",
"f.write(\"Previous Frame MSE: %f\" % mse_prev)\n",
"f.close()"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Model MSE:\t 14.119876861572266\n",
"Prev Frame MSE:\t 0.02834348939359188\n"
]
}
],
"source": [
"print(\"Model MSE:\\t {}\\nPrev Frame MSE:\\t {}\".format(mse_model,mse_prev))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
" 63%|███████████████████████████████████████████████████▉ | 19/30 [09:38<05:35, 30.47s/it]"
]
}
],
"source": [
"# Plot some predictions\n",
"aspect_ratio = float(X_hat.shape[3]) / X_hat.shape[2]\n",
"plt.figure(figsize = (nt, 7*2*aspect_ratio))\n",
"gs = gridspec.GridSpec(2*7, nt)\n",
"gs.update(wspace=0., hspace=0.2)\n",
"plot_save_dir = os.path.join(RESULTS_SAVE_DIR, 'prediction_plots/')\n",
"if not os.path.exists(plot_save_dir): os.mkdir(plot_save_dir)\n",
"plot_idx = np.random.permutation(X_test.shape[0])[:n_plot]\n",
"for i in tqdm(plot_idx):\n",
" for t in range(nt):\n",
" for c in range(7):\n",
" plt.subplot(gs[t + c*2*nt])\n",
" plt.imshow(X_test[i,t,:,:,c], interpolation='none')\n",
" plt.tick_params(axis='both', which='both', bottom='off', top='off', left='off', right='off', labelbottom='off', labelleft='off')\n",
" if t==0: plt.ylabel('Actual', fontsize=10)\n",
"\n",
" plt.subplot(gs[t + (c*2+1)*nt])\n",
" plt.imshow(X_hat[i,t,:,:,c], interpolation='none')\n",
" plt.tick_params(axis='both', which='both', bottom='off', top='off', left='off', right='off', labelbottom='off', labelleft='off')\n",
" if t==0: plt.ylabel('Predicted', fontsize=10)\n",
"\n",
" plt.savefig(plot_save_dir + 'plot_' + str(i) + '.png')\n",
" plt.clf()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"fig=plt.figure(figsize=(15,10))\n",
"columns = 3\n",
"rows = 4\n",
"for i in range(1,columns+rows +1):\n",
" fig.add_subplot(rows,columns,i)\n",
" plt.imshow(X_test[0,0,:,:,i-1],X_hat[0,0,:,:,i-1])"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"X_hat[0][0][0][0][2]"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.6.4"
}
},
"nbformat": 4,
"nbformat_minor": 2
}

68
Project Final/Evaluation/data_utils.py Normal file
View File

@@ -0,0 +1,68 @@
import hickle as hkl
import numpy as np
from keras import backend as K
from keras.preprocessing.image import Iterator
# Data generator that creates sequences for input into PredNet.
class SequenceGenerator(Iterator):
def __init__(self, data_file, source_file, nt,
batch_size=8, shuffle=False, seed=None,
output_mode='error', sequence_start_mode='all', N_seq=None,
data_format=K.image_data_format()):
self.X = hkl.load(data_file) # X will be like (n_images, nb_cols, nb_rows, nb_channels)
self.sources = hkl.load(source_file) # source for each image so when creating sequences can assure that consecutive frames are from same video
self.nt = nt
self.batch_size = batch_size
self.data_format = data_format
assert sequence_start_mode in {'all', 'unique'}, 'sequence_start_mode must be in {all, unique}'
self.sequence_start_mode = sequence_start_mode
assert output_mode in {'error', 'prediction'}, 'output_mode must be in {error, prediction}'
self.output_mode = output_mode
if self.data_format == 'channels_first':
self.X = np.transpose(self.X, (0, 3, 1, 2))
self.im_shape = self.X[0].shape
if self.sequence_start_mode == 'all': # allow for any possible sequence, starting from any frame
self.possible_starts = np.array([i for i in range(self.X.shape[0] - self.nt) if self.sources[i] == self.sources[i + self.nt - 1]])
elif self.sequence_start_mode == 'unique': #create sequences where each unique frame is in at most one sequence
curr_location = 0
possible_starts = []
while curr_location < self.X.shape[0] - self.nt + 1:
if self.sources[curr_location] == self.sources[curr_location + self.nt - 1]:
possible_starts.append(curr_location)
curr_location += self.nt
else:
curr_location += 1
self.possible_starts = possible_starts
if shuffle:
self.possible_starts = np.random.permutation(self.possible_starts)
if N_seq is not None and len(self.possible_starts) > N_seq: # select a subset of sequences if want to
self.possible_starts = self.possible_starts[:N_seq]
self.N_sequences = len(self.possible_starts)
super(SequenceGenerator, self).__init__(len(self.possible_starts), batch_size, shuffle, seed)
def next(self):
with self.lock:
index_array = next(self.index_generator)
current_index = index_array[0]
current_batch_size = len(index_array)
batch_x = np.zeros((current_batch_size, self.nt) + self.im_shape, np.float32)
for i, idx in enumerate(index_array):
idx = self.possible_starts[idx]
batch_x[i] = self.preprocess(self.X[idx:idx+self.nt])
if self.output_mode == 'error': # model outputs errors, so y should be zeros
batch_y = np.zeros(current_batch_size, np.float32)
elif self.output_mode == 'prediction': # output actual pixels
batch_y = batch_x
return batch_x, batch_y
def preprocess(self, X):
return X.astype(np.float32) / 255
def create_all(self):
X_all = np.zeros((self.N_sequences, self.nt) + self.im_shape, np.float32)
for i, idx in enumerate(self.possible_starts):
X_all[i] = self.preprocess(self.X[idx:idx+self.nt])
return X_all

58
Project Final/Evaluation/keras_utils.py Normal file
View File

@@ -0,0 +1,58 @@
import os
import numpy as np
from keras import backend as K
from keras.legacy.interfaces import generate_legacy_interface, recurrent_args_preprocessor
from keras.models import model_from_json
legacy_prednet_support = generate_legacy_interface(
allowed_positional_args=['stack_sizes', 'R_stack_sizes',
'A_filt_sizes', 'Ahat_filt_sizes', 'R_filt_sizes'],
conversions=[('dim_ordering', 'data_format'),
('consume_less', 'implementation')],
value_conversions={'dim_ordering': {'tf': 'channels_last',
'th': 'channels_first',
'default': None},
'consume_less': {'cpu': 0,
'mem': 1,
'gpu': 2}},
preprocessor=recurrent_args_preprocessor)
# Convert old Keras (1.2) json models and weights to Keras 2.0
def convert_model_to_keras2(old_json_file, old_weights_file, new_json_file, new_weights_file):
from prednet import PredNet
# If using tensorflow, it doesn't allow you to load the old weights.
if K.backend() != 'theano':
os.environ['KERAS_BACKEND'] = backend
reload(K)
f = open(old_json_file, 'r')
json_string = f.read()
f.close()
model = model_from_json(json_string, custom_objects = {'PredNet': PredNet})
model.load_weights(old_weights_file)
weights = model.layers[1].get_weights()
if weights[0].shape[0] == model.layers[1].stack_sizes[1]:
for i, w in enumerate(weights):
if w.ndim == 4:
weights[i] = np.transpose(w, (2, 3, 1, 0))
model.set_weights(weights)
model.save_weights(new_weights_file)
json_string = model.to_json()
with open(new_json_file, "w") as f:
f.write(json_string)
if __name__ == '__main__':
old_dir = './model_data/'
new_dir = './model_data_keras2/'
if not os.path.exists(new_dir):
os.mkdir(new_dir)
for w_tag in ['', '-Lall', '-extrapfinetuned']:
m_tag = '' if w_tag == '-Lall' else w_tag
convert_model_to_keras2(old_dir + 'prednet_kitti_model' + m_tag + '.json',
old_dir + 'prednet_kitti_weights' + w_tag + '.hdf5',
new_dir + 'prednet_kitti_model' + m_tag + '.json',
new_dir + 'prednet_kitti_weights' + w_tag + '.hdf5')

311
Project Final/Evaluation/prednet.py Normal file
View File

@@ -0,0 +1,311 @@
import numpy as np
from keras import backend as K
from keras import activations
from keras.layers import Recurrent
from keras.layers import Conv2D, UpSampling2D, MaxPooling2D
from keras.engine import InputSpec
from keras_utils import legacy_prednet_support
class PredNet(Recurrent):
'''PredNet architecture - Lotter 2016.
Stacked convolutional LSTM inspired by predictive coding principles.
# Arguments
stack_sizes: number of channels in targets (A) and predictions (Ahat) in each layer of the architecture.
Length is the number of layers in the architecture.
First element is the number of channels in the input.
Ex. (3, 16, 32) would correspond to a 3 layer architecture that takes in RGB images and has 16 and 32
channels in the second and third layers, respectively.
R_stack_sizes: number of channels in the representation (R) modules.
Length must equal length of stack_sizes, but the number of channels per layer can be different.
A_filt_sizes: filter sizes for the target (A) modules.
Has length of 1 - len(stack_sizes).
Ex. (3, 3) would mean that targets for layers 2 and 3 are computed by a 3x3 convolution of the errors (E)
from the layer below (followed by max-pooling)
Ahat_filt_sizes: filter sizes for the prediction (Ahat) modules.
Has length equal to length of stack_sizes.
Ex. (3, 3, 3) would mean that the predictions for each layer are computed by a 3x3 convolution of the
representation (R) modules at each layer.
R_filt_sizes: filter sizes for the representation (R) modules.
Has length equal to length of stack_sizes.
Corresponds to the filter sizes for all convolutions in the LSTM.
pixel_max: the maximum pixel value.
Used to clip the pixel-layer prediction.
error_activation: activation function for the error (E) units.
A_activation: activation function for the target (A) and prediction (A_hat) units.
LSTM_activation: activation function for the cell and hidden states of the LSTM.
LSTM_inner_activation: activation function for the gates in the LSTM.
output_mode: either 'error', 'prediction', 'all' or layer specification (ex. R2, see below).
Controls what is outputted by the PredNet.
If 'error', the mean response of the error (E) units of each layer will be outputted.
That is, the output shape will be (batch_size, nb_layers).
If 'prediction', the frame prediction will be outputted.
If 'all', the output will be the frame prediction concatenated with the mean layer errors.
The frame prediction is flattened before concatenation.
Nomenclature of 'all' is kept for backwards compatibility, but should not be confused with returning all of the layers of the model
For returning the features of a particular layer, output_mode should be of the form unit_type + layer_number.
For instance, to return the features of the LSTM "representational" units in the lowest layer, output_mode should be specificied as 'R0'.
The possible unit types are 'R', 'Ahat', 'A', and 'E' corresponding to the 'representation', 'prediction', 'target', and 'error' units respectively.
extrap_start_time: time step for which model will start extrapolating.
Starting at this time step, the prediction from the previous time step will be treated as the "actual"
data_format: 'channels_first' or 'channels_last'.
It defaults to the `image_data_format` value found in your
Keras config file at `~/.keras/keras.json`.
# References
- [Deep predictive coding networks for video prediction and unsupervised learning](https://arxiv.org/abs/1605.08104)
- [Long short-term memory](http://deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdf)
- [Convolutional LSTM network: a machine learning approach for precipitation nowcasting](http://arxiv.org/abs/1506.04214)
- [Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects](http://www.nature.com/neuro/journal/v2/n1/pdf/nn0199_79.pdf)
'''
@legacy_prednet_support
def __init__(self, stack_sizes, R_stack_sizes,
A_filt_sizes, Ahat_filt_sizes, R_filt_sizes,
pixel_max=1., error_activation='relu', A_activation='relu',
LSTM_activation='tanh', LSTM_inner_activation='hard_sigmoid',
output_mode='error', extrap_start_time=None,
data_format=K.image_data_format(), **kwargs):
self.stack_sizes = stack_sizes
self.nb_layers = len(stack_sizes)
assert len(R_stack_sizes) == self.nb_layers, 'len(R_stack_sizes) must equal len(stack_sizes)'
self.R_stack_sizes = R_stack_sizes
assert len(A_filt_sizes) == (self.nb_layers - 1), 'len(A_filt_sizes) must equal len(stack_sizes) - 1'
self.A_filt_sizes = A_filt_sizes
assert len(Ahat_filt_sizes) == self.nb_layers, 'len(Ahat_filt_sizes) must equal len(stack_sizes)'
self.Ahat_filt_sizes = Ahat_filt_sizes
assert len(R_filt_sizes) == (self.nb_layers), 'len(R_filt_sizes) must equal len(stack_sizes)'
self.R_filt_sizes = R_filt_sizes
self.pixel_max = pixel_max
self.error_activation = activations.get(error_activation)
self.A_activation = activations.get(A_activation)
self.LSTM_activation = activations.get(LSTM_activation)
self.LSTM_inner_activation = activations.get(LSTM_inner_activation)
default_output_modes = ['prediction', 'error', 'all']
layer_output_modes = [layer + str(n) for n in range(self.nb_layers) for layer in ['R', 'E', 'A', 'Ahat']]
assert output_mode in default_output_modes + layer_output_modes, 'Invalid output_mode: ' + str(output_mode)
self.output_mode = output_mode
if self.output_mode in layer_output_modes:
self.output_layer_type = self.output_mode[:-1]
self.output_layer_num = int(self.output_mode[-1])
else:
self.output_layer_type = None
self.output_layer_num = None
self.extrap_start_time = extrap_start_time
assert data_format in {'channels_last', 'channels_first'}, 'data_format must be in {channels_last, channels_first}'
self.data_format = data_format
self.channel_axis = -3 if data_format == 'channels_first' else -1
self.row_axis = -2 if data_format == 'channels_first' else -3
self.column_axis = -1 if data_format == 'channels_first' else -2
super(PredNet, self).__init__(**kwargs)
self.input_spec = [InputSpec(ndim=5)]
def compute_output_shape(self, input_shape):
if self.output_mode == 'prediction':
out_shape = input_shape[2:]
elif self.output_mode == 'error':
out_shape = (self.nb_layers,)
elif self.output_mode == 'all':
out_shape = (np.prod(input_shape[2:]) + self.nb_layers,)
else:
stack_str = 'R_stack_sizes' if self.output_layer_type == 'R' else 'stack_sizes'
stack_mult = 2 if self.output_layer_type == 'E' else 1
out_stack_size = stack_mult * getattr(self, stack_str)[self.output_layer_num]
out_nb_row = input_shape[self.row_axis] / 2**self.output_layer_num
out_nb_col = input_shape[self.column_axis] / 2**self.output_layer_num
if self.data_format == 'channels_first':
out_shape = (out_stack_size, out_nb_row, out_nb_col)
else:
out_shape = (out_nb_row, out_nb_col, out_stack_size)
if self.return_sequences:
return (input_shape[0], input_shape[1]) + out_shape
else:
return (input_shape[0],) + out_shape
def get_initial_state(self, x):
input_shape = self.input_spec[0].shape
init_nb_row = input_shape[self.row_axis]
init_nb_col = input_shape[self.column_axis]
base_initial_state = K.zeros_like(x) # (samples, timesteps) + image_shape
non_channel_axis = -1 if self.data_format == 'channels_first' else -2
for _ in range(2):
base_initial_state = K.sum(base_initial_state, axis=non_channel_axis)
base_initial_state = K.sum(base_initial_state, axis=1) # (samples, nb_channels)
initial_states = []
states_to_pass = ['r', 'c', 'e']
nlayers_to_pass = {u: self.nb_layers for u in states_to_pass}
if self.extrap_start_time is not None:
states_to_pass.append('ahat') # pass prediction in states so can use as actual for t+1 when extrapolating
nlayers_to_pass['ahat'] = 1
for u in states_to_pass:
for l in range(nlayers_to_pass[u]):
ds_factor = 2 ** l
nb_row = init_nb_row // ds_factor
nb_col = init_nb_col // ds_factor
if u in ['r', 'c']:
stack_size = self.R_stack_sizes[l]
elif u == 'e':
stack_size = 2 * self.stack_sizes[l]
elif u == 'ahat':
stack_size = self.stack_sizes[l]
output_size = stack_size * nb_row * nb_col # flattened size
reducer = K.zeros((input_shape[self.channel_axis], output_size)) # (nb_channels, output_size)
initial_state = K.dot(base_initial_state, reducer) # (samples, output_size)
if self.data_format == 'channels_first':
output_shp = (-1, stack_size, nb_row, nb_col)
else:
output_shp = (-1, nb_row, nb_col, stack_size)
initial_state = K.reshape(initial_state, output_shp)
initial_states += [initial_state]
if K._BACKEND == 'theano':
from theano import tensor as T
# There is a known issue in the Theano scan op when dealing with inputs whose shape is 1 along a dimension.
# In our case, this is a problem when training on grayscale images, and the below line fixes it.
initial_states = [T.unbroadcast(init_state, 0, 1) for init_state in initial_states]
if self.extrap_start_time is not None:
initial_states += [K.variable(0, int if K.backend() != 'tensorflow' else 'int32')] # the last state will correspond to the current timestep
return initial_states
def build(self, input_shape):
self.input_spec = [InputSpec(shape=input_shape)]
self.conv_layers = {c: [] for c in ['i', 'f', 'c', 'o', 'a', 'ahat']}
for l in range(self.nb_layers):
for c in ['i', 'f', 'c', 'o']:
act = self.LSTM_activation if c == 'c' else self.LSTM_inner_activation
self.conv_layers[c].append(Conv2D(self.R_stack_sizes[l], self.R_filt_sizes[l], padding='same', activation=act, data_format=self.data_format))
act = 'relu' if l == 0 else self.A_activation
self.conv_layers['ahat'].append(Conv2D(self.stack_sizes[l], self.Ahat_filt_sizes[l], padding='same', activation=act, data_format=self.data_format))
if l < self.nb_layers - 1:
self.conv_layers['a'].append(Conv2D(self.stack_sizes[l+1], self.A_filt_sizes[l], padding='same', activation=self.A_activation, data_format=self.data_format))
self.upsample = UpSampling2D(data_format=self.data_format)
self.pool = MaxPooling2D(data_format=self.data_format)
self.trainable_weights = []
nb_row, nb_col = (input_shape[-2], input_shape[-1]) if self.data_format == 'channels_first' else (input_shape[-3], input_shape[-2])
for c in sorted(self.conv_layers.keys()):
for l in range(len(self.conv_layers[c])):
ds_factor = 2 ** l
if c == 'ahat':
nb_channels = self.R_stack_sizes[l]
elif c == 'a':
nb_channels = 2 * self.R_stack_sizes[l]
else:
nb_channels = self.stack_sizes[l] * 2 + self.R_stack_sizes[l]
if l < self.nb_layers - 1:
nb_channels += self.R_stack_sizes[l+1]
in_shape = (input_shape[0], nb_channels, nb_row // ds_factor, nb_col // ds_factor)
if self.data_format == 'channels_last': in_shape = (in_shape[0], in_shape[2], in_shape[3], in_shape[1])
with K.name_scope('layer_' + c + '_' + str(l)):
self.conv_layers[c][l].build(in_shape)
self.trainable_weights += self.conv_layers[c][l].trainable_weights
self.states = [None] * self.nb_layers*3
if self.extrap_start_time is not None:
self.t_extrap = K.variable(self.extrap_start_time, int if K.backend() != 'tensorflow' else 'int32')
self.states += [None] * 2 # [previous frame prediction, timestep]
def step(self, a, states):
r_tm1 = states[:self.nb_layers]
c_tm1 = states[self.nb_layers:2*self.nb_layers]
e_tm1 = states[2*self.nb_layers:3*self.nb_layers]
if self.extrap_start_time is not None:
t = states[-1]
a = K.switch(t >= self.t_extrap, states[-2], a) # if past self.extrap_start_time, the previous prediction will be treated as the actual
c = []
r = []
e = []
# Update R units starting from the top
for l in reversed(range(self.nb_layers)):
inputs = [r_tm1[l], e_tm1[l]]
if l < self.nb_layers - 1:
inputs.append(r_up)
inputs = K.concatenate(inputs, axis=self.channel_axis)
i = self.conv_layers['i'][l].call(inputs)
f = self.conv_layers['f'][l].call(inputs)
o = self.conv_layers['o'][l].call(inputs)
_c = f * c_tm1[l] + i * self.conv_layers['c'][l].call(inputs)
_r = o * self.LSTM_activation(_c)
c.insert(0, _c)
r.insert(0, _r)
if l > 0:
r_up = self.upsample.call(_r)
# Update feedforward path starting from the bottom
for l in range(self.nb_layers):
ahat = self.conv_layers['ahat'][l].call(r[l])
if l == 0:
ahat = K.minimum(ahat, self.pixel_max)
frame_prediction = ahat
# compute errors
e_up = self.error_activation(ahat - a)
e_down = self.error_activation(a - ahat)
e.append(K.concatenate((e_up, e_down), axis=self.channel_axis))
if self.output_layer_num == l:
if self.output_layer_type == 'A':
output = a
elif self.output_layer_type == 'Ahat':
output = ahat
elif self.output_layer_type == 'R':
output = r[l]
elif self.output_layer_type == 'E':
output = e[l]
if l < self.nb_layers - 1:
a = self.conv_layers['a'][l].call(e[l])
a = self.pool.call(a) # target for next layer
if self.output_layer_type is None:
if self.output_mode == 'prediction':
output = frame_prediction
else:
for l in range(self.nb_layers):
layer_error = K.mean(K.batch_flatten(e[l]), axis=-1, keepdims=True)
all_error = layer_error if l == 0 else K.concatenate((all_error, layer_error), axis=-1)
if self.output_mode == 'error':
output = all_error
else:
output = K.concatenate((K.batch_flatten(frame_prediction), all_error), axis=-1)
states = r + c + e
if self.extrap_start_time is not None:
states += [frame_prediction, t + 1]
return output, states
def get_config(self):
config = {'stack_sizes': self.stack_sizes,
'R_stack_sizes': self.R_stack_sizes,
'A_filt_sizes': self.A_filt_sizes,
'Ahat_filt_sizes': self.Ahat_filt_sizes,
'R_filt_sizes': self.R_filt_sizes,
'pixel_max': self.pixel_max,
'error_activation': self.error_activation.__name__,
'A_activation': self.A_activation.__name__,
'LSTM_activation': self.LSTM_activation.__name__,
'LSTM_inner_activation': self.LSTM_inner_activation.__name__,
'data_format': self.data_format,
'extrap_start_time': self.extrap_start_time,
'output_mode': self.output_mode}
base_config = super(PredNet, self).get_config()
return dict(list(base_config.items()) + list(config.items()))