double Trainer::PreviousMinibatchEvaluationAverage() const
{
if (!m_evaluationFunction)
InvalidArgument("Trainer::PreviousMinibatchEvaluationAverage: Cannot get evaluation criterion value when no evaluation function was specified during 'this' trainer's construction");
return (GetScalarValue(m_prevMinibatchAggregateEvalCriterionValue) / m_prevMinibatchNumSamples);
}
void CrawlStack(lua_State* L,LuaStack& S,int level)
{
int t = lua_type(L, level);
if (t==LUA_TTABLE) {
TableCrawler C(L,level);
boost::shared_ptr<LuaTable> v=C.GetTable();
S.Append(v);
} else {
S.Append(GetScalarValue(L,level));
}
}
double Trainer::TestMinibatch(const std::unordered_map<Variable, ValuePtr>& arguments, const DeviceDescriptor& computeDevice /*= DeviceDescriptor::UseDefaultDevice()*/)
{
if (!m_aggregatedEvaluationFunction)
InvalidArgument("Trainer::TestMinibatch: Cannot test when no evaluation function was specified during 'this' trainer's construction");
// TODO: Should we refactor this code that is somewhat similar to the prologue of the TrainMinibatch function
std::unordered_map<Variable, ValuePtr> outputs = { { m_aggregatedEvaluationFunction, nullptr }, { m_testSampleCountVar, nullptr } };
m_combinedTrainingFunction->Forward(arguments, outputs, computeDevice);
auto sampleCount = GetSampleCount(m_testSampleCountVar, outputs[m_testSampleCountVar]);
return (GetScalarValue(outputs[m_aggregatedEvaluationFunction]) / sampleCount);
}
void CrawlStack(lua_State* L,LuaStack& S)
{
int top = lua_gettop(L);
for (int i = 1; i <= top; i++) { /* repeat for each level */
int t = lua_type(L, i);
if (t==LUA_TTABLE) {
TableCrawler C(L,i);
boost::shared_ptr<LuaTable> v=C.GetTable();
S.Append(v);
} else {
S.Append(GetScalarValue(L,i));
}
}
}
double Trainer::PreviousMinibatchLossAverage() const
{
return (GetScalarValue(m_prevMinibatchAggregateTrainingLossValue) / m_prevMinibatchNumSamples);
}
