def build_and_save_model(self, encoded_chord_string,
encoded_duration_string):
self.chordsmc = mc.MarkovChain().from_data(encoded_chord_string)
with open('ml_models/markov_bass_chords',
'wb') as outp: # Overwrites any existing file.
pickle.dump(self.chordsmc, outp, pickle.HIGHEST_PROTOCOL)
self.durationmc = mc.MarkovChain().from_data(encoded_duration_string)
with open('ml_models/markov_bass_durations',
'wb') as outp: # Overwrites any existing file.
pickle.dump(self.durationmc, outp, pickle.HIGHEST_PROTOCOL)
def forecast_markov(trainingdata, evaldata):
n = len(evaldata)
mod = mc.MarkovChain().from_data(trainingdata['count'])
ids, states = mod.simulate(n, start=10)
plot(f'markov_{n}.png', states, 'Forecast based on Markov Chains',
evaldata['count'].values)
print('Accuracy Scores for Markov Chains:')
accuracy.eval_model(states, 1, evaldata['count'].values)
def predict(prev_play, oppnent_history, memoryLength):
if len(oppnent_history) > memoryLength:
oppnent_history.pop(0)
chain = mc.MarkovChain().from_data(oppnent_history)
predictionNextItem = giveMostProbableNextItem(chain, prev_play)
winningMove = winDict[predictionNextItem]
return winningMove
def getMarkovProbas(observations):
markovChain = mc.MarkovChain().from_data(observations)
# ~ print (observations)
print(
pandas.DataFrame(markovChain.observed_p_matrix,
index=markovChain.states,
columns=markovChain.states,
dtype=float))
return markovChain.observed_p_matrix
def listen():
global keyboardListener, observations, playBuffer, keyboardController, lastKeyPlayed, period, playThreeshold, lastKeyPressedTime, silentEnding, silentEndingRemoved
keyboardListener = keyboard.Listener(on_press=on_press,
on_release=on_release)
keyboardListener.start()
print("starting listening to keyboard events")
while True:
timeElapsedSinceKeyboard = datetime.now() - lastKeyPressedTime
if timeElapsedSinceKeyboard.seconds > playThreeshold: # play mode, uninterrupted
if len(playBuffer) == 0: # nothing to play yet
if not silentEndingRemoved:
observations = observations[:-silentEnding]
silentEndingRemoved = True
if len(observations) > 500:
# ~ print(observations)
markovChain = mc.MarkovChain().from_data(observations)
samplesToPlay = min(
int(len(observations) / 2), int(10000 / period)
) # will play longer on stronger databases, 10s max
# ~ ids, playBuffer = markovChain.simulate(samplesToPlay, start=lastKeyPressed)
ids, playBuffer = markovChain.simulate(
10000, start=lastKeyPressed)
else: # we are continuing to play previously computed predictions
if not continuousLearning and len(observations) > 0:
observations = numpy.array(
[]) #clear database when playing
if playBuffer[0] == "0": # this is a pause
print("PAUSE")
if lastKeyPlayed is not None: # we've pressed a key
keyboardController.release(lastKeyPlayed)
print("OFF_" + lastKeyPlayed)
lastKeyPlayed = None
elif playBuffer[
0] != lastKeyPlayed and lastKeyPlayed is not None: # were changing from one key to another
keyboardController.release(lastKeyPlayed)
print("OFF_" + lastKeyPlayed)
lastKeyPlayed = playBuffer[0]
keyboardController.press(playBuffer[0])
print("ON_" + lastKeyPlayed)
elif lastKeyPlayed is None: # we are pressing a key coming from a pause
lastKeyPlayed = playBuffer[0]
keyboardController.press(playBuffer[0])
print("ON_" + lastKeyPlayed)
playBuffer = numpy.delete(playBuffer, 0)
elif len(keysCurrentlyPressed) > 0: # record mode : a key is pressed
observations = numpy.append(observations, lastKeyPressed)
if len(playBuffer) > 0: playBuffer = numpy.array([])
else: # record mode : nothing is pressed
observations = numpy.append(observations, '0')
if len(playBuffer) > 0: time.sleep(period / 1000)
else: time.sleep(period / 1000)
def getMarkovProbas(observations) :
global removeSilenceFromVisualisation
if removeSilenceFromVisualisation : observations = observations[observations != "0"]
markovChain = mc.MarkovChain().from_data(observations)
names = [n for n in markovChain.states] # needed to convert np array into JSON-serialisable list
weights = markovChain.observed_p_matrix.tolist()
# ~ weights = [[round(w, 2) for w in W] for W in weights] # round floats to 3 digits
weights = [[w for w in W] for W in weights]
# ~ print("states :", names)
# ~ print("probas :", weights)
if len(names) > 1 : return {"weights":weights, "names":names}
else : return False
def getMarkovProbas(observations):
global keyToNote, removeSilenceFromVisualisation
if removeSilenceFromVisualisation:
observations = observations[observations != "0"]
markovChain = mc.MarkovChain().from_data(observations)
names = markovChain.states
weights = markovChain.observed_p_matrix.tolist()
# ~ weights = [[round(w, 2) for w in W] for W in weights] # round floats to 3 digits
weights = [[w for w in W] for W in weights]
names = [keyToNote[n] for n in names
] # translate keystrokes into corresponding readable note names
print("states :", names)
print("probas :", weights)
# ~ print(pandas.DataFrame(markovChain.observed_p_matrix, index=markovChain.states, columns=markovChain.states, dtype=float))
if len(names) > 1: return {"weights": weights, "names": names}
else: return False
#%%
import mchmm as mc
#%%
chain = mc.MarkovChain().from_data('RSPPSRRRSSSSPRSSPRSSRPRPPSR')
# %%
chain.observed_matrix
# %%
graph = chain.graph_make(format="png",
graph_attr=[("rankdir", "LR")],
node_attr=[("fontname", "Roboto bold"),
("fontsize", "20")],
edge_attr=[("fontname", "Iosevka"),
("fontsize", "12")])
graph.render()
# %%
# %%
return markovChain.observed_p_matrix
if __name__ == '__main__':
keyboardListener = keyboard.Listener(on_press=on_press,
on_release=on_release)
keyboardListener.start()
while True:
timeElapsedSinceKeyboard = datetime.now() - lastKeyPressedTime
if timeElapsedSinceKeyboard.seconds > playThreeshold: # play mode, uninterrupted
if len(playBuffer) == 0: # nothing to play yet
if len(observations) > 500:
print(observations)
markovChain = mc.MarkovChain().from_data(observations)
ids, playBuffer = markovChain.simulate(
10000, start=lastKeyPressed)
else: # we are continuing to play previously computed predictions
if playBuffer[0] == "0": # this is a pause
print("PAUSE")
if lastKeyPlayed is not None: # we've pressed a key
keyboardController.release(lastKeyPlayed)
print("OFF_" + lastKeyPlayed)
lastKeyPlayed = None
elif playBuffer[
0] != lastKeyPlayed and lastKeyPlayed is not None: # were changing from one key to another
keyboardController.release(lastKeyPlayed)
print("OFF_" + lastKeyPlayed)
lastKeyPlayed = playBuffer[0]
keyboardController.press(playBuffer[0])
# Slicing our requests
searchPath = searchPath + '&from=' + str(begIndex) + '&size=' + str(batchSize)
response = requests.get(CONSUMER_URL + searchPath, headers=headers, verify=False)
# I: https://docs.python.org/3/library/http.client.html#httpresponse-objects
respData = response.json()
observationsArray = respData["hits"]["hits"]
for observation in observationsArray:
# Important to convert to seconds, datetime cannot handle ms
# TODO: consider rounding to minutes
timestamp_s = int(observation['_source']['timestamp'] / 1000)
count = int(observation['_source']['value'])
cur_date = datetime.fromtimestamp(timestamp_s)
df_row = pd.DataFrame([[cur_date, count]], columns=['t', 'count'])
retrievedData = retrievedData.append(df_row)
begIndex = begIndex + batchSize
countLeft = countLeft - batchSize
print(retrievedData)
retrievedData.index = retrievedData.t
# Create the model from scratch
mod = mc.MarkovChain().from_data(retrievedData['count'])
print(mod.expected_matrix)
pickle.dump(mod, open(modelFilepath, 'wb'))
else:
print("Not enough arguments. The call should be: python3 markovderivator.py <sensor id> <model file name> <number of weeks|0> <batch size>")
