Start values as results of previous simulation - modelica

Is it possible to use the result of a simulation Sim 1 at time t as start value of simulation Sim 2? The use of extend doesn't work for start values.
Example:
model Sim 1
Real a;
equation
a=2*time;
end Sim 1;
for model Sim 2, I need
Real b (start=a at time t)
to use in several other set of equations.

You have to differ between the modeling and the simulation process:
With the language Modelica you define your models
With the simulation tool (like Dymola) you perform the simulation.
The keyword extends is part of the Modelica language. So it cannot be of any use in this context, as you use it to define models, not to describe how a simulation should be performed.
The solution for your problem must be searched in the simulation tool and Dymola offers a simulator function, which does exactly what you want: simulateExtendedModel. It allows to read the final value of a variable and you can initialize parameters and state variables with it. You can use it in a .mos script or within a Modelica function.
So if we rename your models Sim1 and Sim2 to Model1 and Model2 (because they are really models, not simulations) the function below would do what you want:
function sim
import DymolaCommands.SimulatorAPI.simulateExtendedModel;
protected
Boolean ok;
Real a;
Real[1] finalValues;
algorithm
(ok, finalValues) :=simulateExtendedModel("Model1", 0, 5, finalNames={"a"});
a :=finalValues[1];
simulateExtendedModel("Model2", 5, 10, initialNames={"b"}, initialValues={a});
end sim;
If you want to set multiple variables, you can use this code:
function sim2
import DymolaCommands.SimulatorAPI.simulateExtendedModel;
protected
Boolean ok;
Real[:] finalValues_sim1;
String[:] finalNames_sim1 = {"a1", "a2", "a3"};
String[:] initialNames_sim2 = {"b1", "b2", "b3"};
algorithm
(ok, finalValues_sim1) :=simulateExtendedModel("SO.Model1", 0, 5, finalNames=finalNames_sim1);
simulateExtendedModel("SO.Model2", 5, 10, initialNames=initialNames_sim2, initialValues=finalValues_sim1);
end sim2;

Related

Is it possible to use callbacks to access a single trajectory in Julia's DifferentialEquations Ensemble Problems?

I am new to Julia and trying to use the Julia package DifferentialEquations to simultaneously solve for several conditions of the same set of coupled ODEs. My system is a model of an experiment and in one of the conditions, I increase the amount of one of the dependent variables at mid-way through the process.
I would like to be able to adjust the condition of this single trajectory, however so far I am only able to adjust all the trajectories at once. Is it possible to access a single one using callbacks? If not, is there a better way to do this?
Here is a simplified example using the lorentz equations for what I want to be doing:
#Differential Equations setup
function lorentz!(du,u,p,t)
a,r,b=p
du[1]= a*(u[2]-u[1])
du[2]=u[1]*(r-u[3])-u[2]
du[3]=u[1]*u[2]-b*u[3];
end
#function to cycle through inital conditions
function prob_func(prob,i,repeat)
remake(prob; u0 = u0_arr[i]);
end
#inputs
t_span=[(0.0,100.0),(0.0,100.0)];
u01=[0.0;1.0;0.0];
u02=[0.0;1.0;0.0];
u0_arr = [u01,u02];
p=[10.,28.,8/3];
#initialising the Ensemble Problem
prob = ODEProblem(lorentz!,u0_arr[1],t_span[1],p);
CombinedProblem = EnsembleProblem(prob,
prob_func = prob_func, #-> (prob),#repeat is a count for how many times the trajectories had been repeated
safetycopy = true # determines whether a safetly deepcopy is called on the prob before the prob_func (sounds best to leave as true for user-given prob_func)
);
#introducing callback
function condition(u,t,repeat)
return 50 .-t
end
function affect!(repeat)
repeat.u[1]=repeat.u[1] +50
end
callback = DifferentialEquations.ContinuousCallback(condition, affect!)
#solving
sim=solve(CombinedProblem,Rosenbrock23(),EnsembleSerial(),trajectories=2,callback=callback);
# Plotting for ease of understanding example
plot(sim[1].t,sim[1][1,:])
plot!(sim[2].t,sim[2][1,:])
I want to produce something like this:
Example_desired_outcome
But this code produces:
Example_current_outcome
Thank you for your help!
You can make that callback dependent on a parameter and make the parameter different between problems. For example:
function f(du,u,p,t)
if p == 0
du[1] = 2u[1]
else
du[1] = -2u[1]
end
du[2] = -u[2]
end
condition(t,u,integrator) = u[2] - 0.5
affect!(integrator) = integrator.prob.p = 1
For more information, check out the FAQ on this topic: https://diffeq.sciml.ai/stable/basics/faq/#Switching-ODE-functions-in-the-middle-of-integration

How to implement exponentially decay learning rate in Keras by following the global steps

Look at the following example
# encoding: utf-8
import numpy as np
import pandas as pd
import random
import math
from keras import Sequential
from keras.layers import Dense, Activation
from keras.optimizers import Adam, RMSprop
from keras.callbacks import LearningRateScheduler
X = [i*0.05 for i in range(100)]
def step_decay(epoch):
initial_lrate = 1.0
drop = 0.5
epochs_drop = 2.0
lrate = initial_lrate * math.pow(drop,
math.floor((1+epoch)/epochs_drop))
return lrate
def build_model():
model = Sequential()
model.add(Dense(32, input_shape=(1,), activation='relu'))
model.add(Dense(1, activation='linear'))
adam = Adam(lr=0.5)
model.compile(loss='mse', optimizer=adam)
return model
model = build_model()
lrate = LearningRateScheduler(step_decay)
callback_list = [lrate]
for ep in range(20):
X_train = np.array(random.sample(X, 10))
y_train = np.sin(X_train)
X_train = np.reshape(X_train, (-1,1))
y_train = np.reshape(y_train, (-1,1))
model.fit(X_train, y_train, batch_size=2, callbacks=callback_list,
epochs=1, verbose=2)
In this example, the LearningRateSchedule does not change the learning rate at all because in each iteration of ep, epoch=1. Thus the learning rate is just const (1.0, according to step_decay). In fact, instead of setting epoch>1 directly, I have to do outer loop as shown in the example, and insider each loop, I just run 1 epoch. (This is the case when I implement deep reinforcement learning, instead of supervised learning).
My question is how to set an exponentially decay learning rate in my example and how to get the learning rate in each iteration of ep.
You can actually pass two arguments to the LearningRateScheduler.
According to Keras documentation, the scheduler is
a function that takes an epoch index as input (integer, indexed from
0) and current learning rate and returns a new learning rate as output
(float).
So, basically, simply replace your initial_lr with a function parameter, like so:
def step_decay(epoch, lr):
# initial_lrate = 1.0 # no longer needed
drop = 0.5
epochs_drop = 2.0
lrate = lr * math.pow(drop,math.floor((1+epoch)/epochs_drop))
return lrate
The actual function you implement is not exponential decay (as you mention in your title) but a staircase function.
Also, you mention your learning rate does not change inside your loop. That's true because you set model.fit(..., epochs=1,...) and your epochs_drop = 2.0 at the same time. I am not sure this is your desired case or not. You are providing a toy example and it's not clear in that case.
I would like to add the more common case where you don't mix a for loop with fit() and just provide a different epochs parameter in your fit() function. In this case you have the following options:
First of all keras provides a decaying functionality itself with the predefined optimizers. For example in your case Adam() the actual code is:
lr = lr * (1. / (1. + self.decay * K.cast(self.iterations, K.dtype(self.decay))))
which is not exactly exponential either and it's somehow different than tensorflow's one. Also, it's used only when decay > 0.0 as it's obvious.
To follow the tensorflow convention of exponential decay you should implement:
decayed_learning_rate = learning_rate * ^ (global_step / decay_steps)
Depending on your needs you could choose to implement a Callback subclass and define a function within it (see 3rd bullet below) or use LearningRateScheduler which is actually exactly this with some checking: a Callback subclass which updates the learning rate at each epoch end.
If you want a finer handling of your learning rate policy (per batch for example) you would have to implement your subclass since as far as I know there is no implemented subclass for this task. The good part is that it's super easy:
Create a subclass
class LearningRateExponentialDecay(Callback):
and add the __init__() function which will initialize your instance with all needed parameters and also create a global_step variables to keep track of the iterations (batches):
def __init__(self, init_learining_rate, decay_rate, decay_steps):
self.init_learining_rate = init_learining_rate
self.decay_rate = decay_rate
self.decay_steps = decay_steps
self.global_step = 0
Finally, add the actual function inside the class:
def on_batch_begin(self, batch, logs=None):
actual_lr = float(K.get_value(self.model.optimizer.lr))
decayed_learning_rate = actual_lr * self.decay_rate ^ (self.global_step / self.decay_steps)
K.set_value(self.model.optimizer.lr, decayed_learning_rate)
self.global_step += 1
The really cool part is the if you want the above subclass to update every epoch you could use on_epoch_begin(self, epoch, logs=None) which nicely has epoch as parameter to it's signature. This case is even easier as you could skip global step altogether (no need to keep track of it now unless you want a fancier way to apply your decay) and use epoch in it's place.

z3py: How to implement a counter in z3?

I want to design logics similar to a counter in Z3py.
If writing python script, we usually define a variable "counter" and then keep incrementing it when necessary. However, in Z3, there is no variant. Therefore, instead of defining an variant, I define a trace of that variant.
This is a sample code. Suppose there is an array "myArray" of size 5, and the elements in the array are 1 or 2. I want to assert a constraint that there must be two '2's in "myArray"
from z3 import *
s = Solver()
myArray = IntVector('myArray',5)
for i in range(5):
s.add(Or(myArray[i]==1,myArray[i]==2))
counterTrace = IntVector('counterTrace',6)
s.add(counterTrace[0]==0)
for i in range(5):
s.add(If(myArray[i]==2,counterTrace[i+1]==counterTrace[i]+1,counterTrace[i+1]==counterTrace[i]))
s.add(counterTrace[5]==2)
print s.check()
print s.model()
My question is that is this an efficient way of implementing the concept of counter in Z3? In my real problem, which is more complicated, this is really inefficient.
You can do this but it is much easier to create the sum over myArray[i] == 2 ? 1 : 0. That way you don't need to assert anything and you are dealing with normal expressions.

Does a function cache exist in Matlab?

In Python we have lru_cache as a function wrapper. Add it to your function and the function will only be evaluated once per different input argument.
Example (from Python docs):
#lru_cache(maxsize=None)
def fib(n):
if n < 2:
return n
return fib(n-1) + fib(n-2)
>>> [fib(n) for n in range(16)]
[0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610]
>>> fib.cache_info()
CacheInfo(hits=28, misses=16, maxsize=None, currsize=16)
I wonder whether a similar thing exists in Matlab? At the moment I am using cache files, like so:
function result = fib(n):
% FIB example like the Python example. Don't implement it like that!
cachefile = ['fib_', n, '.mat'];
try
load(cachefile);
catch e
if n < 2
result = n;
else
result = fib(n-1) + fib(n-2);
end
save(cachefile, 'result');
end
end
The problem I have with doing it this way, is that if I change my function, I need to delete the cachefile.
Is there a way to do this with Matlab realising when I changed the function and the cache has become invalidated?
Since matlab 2017 this is available:
https://nl.mathworks.com/help/matlab/ref/memoizedfunction.html
a = memoized(#sin)
I've created something like this for my own personal use: a CACHE class. (I haven't documented the code yet though.) It appears to be more flexible than Python's lru_cache (I wasn't aware of that, thanks) in that it has several methods for adjusting exactly what gets cached (to save memory) and how the comparisons are made. It could still use some refinement (#Daniel's suggestion to use the containers.Map class is a good one – though it would limit compatibility with old Matlab versions). The code is on GitHub so you're welcome to fork and improve it.
Here is a basic example of how it can be used:
function Output1 = CacheDemo(Input1,Input2)
persistent DEMO_CACHE
if isempty(DEMO_CACHE)
% Initialize cache object on first run
CACHE_SIZE = 10; % Number of input/output patterns to cache
DEMO_CACHE = CACHE(CACHE_SIZE,Input1,Input2);
CACHE_IDX = 1;
else
% Check if input pattern corresponds something stored in cache
% If not, return next available CACHE_IDX
CACHE_IDX = DEMO_CACHE.IN([],Input1,Input2);
if ~isempty(CACHE_IDX) && DEMO_CACHE.OUT(CACHE_IDX) > 0
[~,Output1] = DEMO_CACHE.OUT(CACHE_IDX);
return;
end
end
% Perform computation
Output1 = rand(Input1,Input2);
% Save output to cache CACHE_IDX
DEMO_CACHE.OUT(CACHE_IDX,Output1);
I created this class to cache the results from time-consuming stochastic simulations and have since used it to good effect in a few other places. If there is interest, I might be willing to spend some time documenting the code sooner as opposed to later. It would be nice if there was a way to limit memory use as well (a big consideration in my own applications), but getting the size of arbitrary Matlab datatypes is not trivial. I like your idea of caching to a file, which might be a good idea for larger data. Also, it might be nice to create a "lite" version that does what Python's lru_cache does.

PyBrain how to interpret the results from net.activate?

I've trained a network on PyBrain for purpose of classification and am ready to fire away with specific input. However, when I do
classes = ['apple', 'orange', 'peach', 'banana']
data = ClassificationDataSet(len(input), 1, nb_classes=len(classes), class_labels=classes)
data._convertToOneOfMany( ) # recommended by PyBrain
fnn = buildNetwork( data.indim, 5, data.outdim, outclass=SoftmaxLayer )
trainer = BackpropTrainer( fnn, dataset=data, momentum=m, verbose=True, weightdecay=wd)
trainer.trainUntilConvergence(maxEpochs=80)
# stop training and start using my trained network here
output = fnn.activate(input)
As expected, I get a numeric value for "output", but is there a way to determine the predicted class label directly? Even if there's not one, how can I map the value of "output" to my class label? Thank you for your help.
When you say you get a numeric value for "output" do you mean a scalar (that is, not an array)? From my understanding of it, you should have gotten an array of four values (ie. as many as possible output classes you have). The biggest value in that array corresponds to the index of the class. I don't know if PyBrain provides an utility function to extract that, but you can do it like this:
class_index = max(xrange(len(output)), key=output.__getitem__)
class_name = classes[class_index]
Incidentally, you omitted the step in which you actually fill the data in the dataset.