Die Migration der Bereiche "Docker Registry" und "Artifiacts" ist fast abgeschlossen. Die letzten Daten werden im Laufe des heutigen Abend (05.08.2021) noch vollständig hochgeladen. Das Anlegen neuer Images und Artifacts funktioniert bereits wieder.

Commit 2157ebad authored by Sebastian N.'s avatar Sebastian N.
Browse files

Added flatten parameter to FullyConnected layer and added bidirectional parameter to RNN layers

parent aaef2c79
Pipeline #173124 passed with stages
in 3 minutes and 25 seconds
<#assign flatten = element.element.inputTypes[0].height != 1 || element.element.inputTypes[0].width != 1>
<#assign input = element.inputs[0]>
<#assign units = element.units?c>
<#assign use_bias = element.noBias?string("False","True")>
<#assign flatten = element.flatten?string("True","False")>
<#if mode == "ARCHITECTURE_DEFINITION">
<#if flatten>
self.${element.name}flatten = gluon.nn.Flatten()
</#if>
self.${element.name} = gluon.nn.Dense(units=${units}, use_bias=${use_bias})
self.${element.name} = gluon.nn.Dense(units=${units}, use_bias=${use_bias}, flatten=${flatten})
<#include "OutputShape.ftl">
<#elseif mode == "FORWARD_FUNCTION">
<#if flatten>
${element.name}flatten_ = self.${element.name}flatten(${input})
<#assign input = element.name + "flatten_">
</#if>
${element.name} = self.${element.name}(${input})
</#if>
\ No newline at end of file
<#if element.member == "NONE">
<#assign input = element.inputs[0]>
<#if mode == "ARCHITECTURE_DEFINITION">
self.${element.name} = gluon.rnn.GRU(hidden_size=${element.units?c}, num_layers=${element.layers?c}, layout='NTC')
self.${element.name} = gluon.rnn.GRU(hidden_size=${element.units?c},
num_layers=${element.layers?c},
bidirectional=${element.bidirectional?string("True", "False")}
layout='NTC')
<#include "OutputShape.ftl">
<#elseif mode == "FORWARD_FUNCTION">
<#if element.isVariable()>
......
<#if element.member == "NONE">
<#assign input = element.inputs[0]>
<#if mode == "ARCHITECTURE_DEFINITION">
self.${element.name} = gluon.rnn.LSTM(hidden_size=${element.units?c}, num_layers=${element.layers?c}, layout='NTC')
self.${element.name} = gluon.rnn.LSTM(hidden_size=${element.units?c},
num_layers=${element.layers?c},
bidirectional=${element.bidirectional?string("True", "False")}
layout='NTC')
<#include "OutputShape.ftl">
<#elseif mode == "FORWARD_FUNCTION">
<#if element.isVariable()>
......
<#if element.member == "NONE">
<#assign input = element.inputs[0]>
<#if mode == "ARCHITECTURE_DEFINITION">
self.${element.name} = gluon.rnn.RNN(hidden_size=${element.units?c}, num_layers=${element.layers?c}, activation='tanh', layout='NTC')
self.${element.name} = gluon.rnn.RNN(hidden_size=${element.units?c},
num_layers=${element.layers?c},
bidirectional=${element.bidirectional?string("True", "False")},
activation='tanh',
layout='NTC')
<#include "OutputShape.ftl">
<#elseif mode == "FORWARD_FUNCTION">
<#if element.isVariable()>
......
......@@ -191,18 +191,17 @@ class Net_0(gluon.HybridBlock):
self.concatenate6_ = Concatenate(dim=1)
# concatenate6_, output shape: {[256,6,6]}
self.fc6_flatten = gluon.nn.Flatten()
self.fc6_ = gluon.nn.Dense(units=4096, use_bias=True)
self.fc6_ = gluon.nn.Dense(units=4096, use_bias=True, flatten=True)
# fc6_, output shape: {[4096,1,1]}
self.relu6_ = gluon.nn.Activation(activation='relu')
self.dropout6_ = gluon.nn.Dropout(rate=0.5)
self.fc7_ = gluon.nn.Dense(units=4096, use_bias=True)
self.fc7_ = gluon.nn.Dense(units=4096, use_bias=True, flatten=True)
# fc7_, output shape: {[4096,1,1]}
self.relu7_ = gluon.nn.Activation(activation='relu')
self.dropout7_ = gluon.nn.Dropout(rate=0.5)
self.fc8_ = gluon.nn.Dense(units=10, use_bias=True)
self.fc8_ = gluon.nn.Dense(units=10, use_bias=True, flatten=True)
# fc8_, output shape: {[10,1,1]}
self.softmax8_ = Softmax()
......@@ -262,8 +261,7 @@ class Net_0(gluon.HybridBlock):
pool5_2_ = self.pool5_2_(conv5_2_)
relu5_2_ = self.relu5_2_(pool5_2_)
concatenate6_ = self.concatenate6_(relu5_1_, relu5_2_)
fc6_flatten_ = self.fc6_flatten(concatenate6_)
fc6_ = self.fc6_(fc6_flatten_)
fc6_ = self.fc6_(concatenate6_)
relu6_ = self.relu6_(fc6_)
dropout6_ = self.dropout6_(relu6_)
fc7_ = self.fc7_(dropout6_)
......
......@@ -349,11 +349,11 @@ class Net_0(gluon.HybridBlock):
self.globalpooling31_ = gluon.nn.GlobalAvgPool2D()
# globalpooling31_, output shape: {[64,1,1]}
self.fc31_ = gluon.nn.Dense(units=128, use_bias=True)
self.fc31_ = gluon.nn.Dense(units=128, use_bias=True, flatten=True)
# fc31_, output shape: {[128,1,1]}
self.dropout31_ = gluon.nn.Dropout(rate=0.5)
self.fc32_ = gluon.nn.Dense(units=10, use_bias=True)
self.fc32_ = gluon.nn.Dense(units=10, use_bias=True, flatten=True)
# fc32_, output shape: {[10,1,1]}
self.softmax32_ = Softmax()
......
......@@ -219,18 +219,17 @@ class Net_0(gluon.HybridBlock):
strides=(2,2))
# pool13_, output shape: {[512,7,7]}
self.fc13_flatten = gluon.nn.Flatten()
self.fc13_ = gluon.nn.Dense(units=4096, use_bias=True)
self.fc13_ = gluon.nn.Dense(units=4096, use_bias=True, flatten=True)
# fc13_, output shape: {[4096,1,1]}
self.relu14_ = gluon.nn.Activation(activation='relu')
self.dropout14_ = gluon.nn.Dropout(rate=0.5)
self.fc14_ = gluon.nn.Dense(units=4096, use_bias=True)
self.fc14_ = gluon.nn.Dense(units=4096, use_bias=True, flatten=True)
# fc14_, output shape: {[4096,1,1]}
self.relu15_ = gluon.nn.Activation(activation='relu')
self.dropout15_ = gluon.nn.Dropout(rate=0.5)
self.fc15_ = gluon.nn.Dense(units=1000, use_bias=True)
self.fc15_ = gluon.nn.Dense(units=1000, use_bias=True, flatten=True)
# fc15_, output shape: {[1000,1,1]}
self.softmax15_ = Softmax()
......@@ -282,8 +281,7 @@ class Net_0(gluon.HybridBlock):
conv13_ = self.conv13_(conv13_padding)
relu13_ = self.relu13_(conv13_)
pool13_ = self.pool13_(relu13_)
fc13_flatten_ = self.fc13_flatten(pool13_)
fc13_ = self.fc13_(fc13_flatten_)
fc13_ = self.fc13_(pool13_)
relu14_ = self.relu14_(fc13_)
dropout14_ = self.dropout14_(relu14_)
fc14_ = self.fc14_(dropout14_)
......
......@@ -90,11 +90,11 @@ class Net_0(gluon.HybridBlock):
else:
self.input_normalization_state_ = NoNormalization()
self.fc2_1_ = gluon.nn.Dense(units=300, use_bias=True)
self.fc2_1_ = gluon.nn.Dense(units=300, use_bias=True, flatten=True)
# fc2_1_, output shape: {[300,1,1]}
self.relu2_1_ = gluon.nn.Activation(activation='relu')
self.fc3_1_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc3_1_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc3_1_, output shape: {[600,1,1]}
if data_mean:
......@@ -104,14 +104,14 @@ class Net_0(gluon.HybridBlock):
else:
self.input_normalization_action_ = NoNormalization()
self.fc2_2_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc2_2_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc2_2_, output shape: {[600,1,1]}
self.fc4_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc4_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc4_, output shape: {[600,1,1]}
self.relu4_ = gluon.nn.Activation(activation='relu')
self.fc5_ = gluon.nn.Dense(units=1, use_bias=True)
self.fc5_ = gluon.nn.Dense(units=1, use_bias=True, flatten=True)
# fc5_, output shape: {[1,1,1]}
......
......@@ -90,11 +90,11 @@ class Net_0(gluon.HybridBlock):
else:
self.input_normalization_state_ = NoNormalization()
self.fc2_1_ = gluon.nn.Dense(units=300, use_bias=True)
self.fc2_1_ = gluon.nn.Dense(units=300, use_bias=True, flatten=True)
# fc2_1_, output shape: {[300,1,1]}
self.relu2_1_ = gluon.nn.Activation(activation='relu')
self.fc3_1_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc3_1_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc3_1_, output shape: {[600,1,1]}
if data_mean:
......@@ -104,14 +104,14 @@ class Net_0(gluon.HybridBlock):
else:
self.input_normalization_action_ = NoNormalization()
self.fc2_2_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc2_2_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc2_2_, output shape: {[600,1,1]}
self.fc4_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc4_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc4_, output shape: {[600,1,1]}
self.relu4_ = gluon.nn.Activation(activation='relu')
self.fc5_ = gluon.nn.Dense(units=1, use_bias=True)
self.fc5_ = gluon.nn.Dense(units=1, use_bias=True, flatten=True)
# fc5_, output shape: {[1,1,1]}
......
......@@ -90,11 +90,11 @@ class Net_0(gluon.HybridBlock):
else:
self.input_normalization_state_ = NoNormalization()
self.fc2_1_ = gluon.nn.Dense(units=300, use_bias=True)
self.fc2_1_ = gluon.nn.Dense(units=300, use_bias=True, flatten=True)
# fc2_1_, output shape: {[300,1,1]}
self.relu2_1_ = gluon.nn.Activation(activation='relu')
self.fc3_1_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc3_1_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc3_1_, output shape: {[600,1,1]}
if data_mean:
......@@ -104,14 +104,14 @@ class Net_0(gluon.HybridBlock):
else:
self.input_normalization_action_ = NoNormalization()
self.fc2_2_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc2_2_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc2_2_, output shape: {[600,1,1]}
self.fc4_ = gluon.nn.Dense(units=600, use_bias=True)
self.fc4_ = gluon.nn.Dense(units=600, use_bias=True, flatten=True)
# fc4_, output shape: {[600,1,1]}
self.relu4_ = gluon.nn.Activation(activation='relu')
self.fc5_ = gluon.nn.Dense(units=1, use_bias=True)
self.fc5_ = gluon.nn.Dense(units=1, use_bias=True, flatten=True)
# fc5_, output shape: {[1,1,1]}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment