def predict():
f = glob.glob('*.d')#.d Es el diccionario que toma un clima y le asigna un número Se genera en
#cuanto procesas los datos
if 'states.d' not in f:
raise "No data has been processed, cannot express any prediction."
with open('states.d', 'rb') as fp:
states = cPickle.load(fp)
with open('test_set.csv', 'rb') as fp:
lines = fp.readlines()
lines = lines[1:] #retorna lista
data = [line[0:line.find('\n')].split(',') for line in lines]
dates = map(lambda x: x[0], data)
data = [f[1:] for f in data]
inputs = np.zeros((len(data[0]), len(data)))
for i,input in enumerate(data): #retorna lista
l = map(lambda x: float(x), input[0:-1])
l.append(states[input[-1]])
inp = np.array([l])
inputs[:, i] = inp
values = linear.predict(inputs) #guarda los datos de prediccion en el formato esperado
with open('output.csv', 'wb') as fp:
fp.write('"fecha", "conteo_ordenes"\n')
for i in range(0, len(dates)):
line = '"'+dates[i]+'"'+','+str(int(np.round(values[i][0])))+'\n'
fp.write(line)
def perceptron_update(bow, label, weights, labels):
update = defaultdict(lambda: defaultdict(float))
prediction, _ = predict(bow, weights, labels)
if prediction == label: # no changes need to be made to the weights
return update
# for mean squared error, `w(t+1) = w(t) + learn_rate * (y - y_pred) * x`
# Here, we are computing `(y - y_pred) * x` as `y * x - y_pred * x`
for word in bow:
update[label][word] += bow[word]
update[prediction][word] -= bow[word]
return update
while True:
test_data, test_events, stopevents = bufhelp.gatherdata(
"stimulus.hybrid",
trlen_ms,
"stimulus.hybrid",
milliseconds=True)
# if isinstance(stopevents, list):
# if any(["stimulus.feedback" in x.type for x in stopevents]):
# break
# else:
# if "stimulus.sequence" in stopevents.type:
# break
test_data = preproc.detrend(test_data)
test_data, badch = preproc.badchannelremoval(test_data)
test_data = preproc.spatialfilter(test_data)
test_data = preproc.spectralfilter(test_data, (1, 10, 15, 25),
bufhelp.fSample)
test_data, test_events, badtrials = preproc.badtrailremoval(
test_data, test_events)
predictions = linear.predict(test_data, classifier)
prediction = str(int(predictions[0]))
bufhelp.sendEvent("classifier.prediction", prediction)
elif e.value == "exit":
run = False
print("Waiting for startPhase.cmd event.")
import preproc
import linear
import pickle
dname ='training_data'
cname ='clsfr'
trlen_ms = 3000
#load the trained classifier
if os.path.exists(cname+'.pk'):
f =pickle.load(open(cname+'.pk','rb'))
classifier = f['classifier']
# connect to the buffer, if no-header wait until valid connection
ftc,hdr=bufhelp.connect()
while True:
# wait for data after a trigger event
data, events, stopevents, state = bufhelp.gatherdata(["stimulus.target"],trlen_ms,[("stimulus.feedback","end")], milliseconds=True)
# YOUR CODE HERE #
# apply classifier, default is a linear-least-squares-classifier
predictions = linear.predict(data)
# send the prediction events
for pred in predictions:
bufhelp.sendEvent("classifier.prediction",pred)
import pickle
dname = 'training_data'
cname = 'clsfr'
trlen_ms = 3000
#load the trained classifier
if os.path.exists(cname + '.pk'):
f = pickle.load(open(cname + '.pk', 'rb'))
classifier = f['classifier']
# connect to the buffer, if no-header wait until valid connection
ftc, hdr = bufhelp.connect()
while True:
# wait for data after a trigger event
data, events, stopevents, state = bufhelp.gatherdata(
["stimulus.target"],
trlen_ms, [("stimulus.feedback", "end")],
milliseconds=True)
# YOUR CODE HERE #
# apply classifier, default is a linear-least-squares-classifier
predictions = linear.predict(data)
# send the prediction events
for pred in predictions:
bufhelp.sendEvent("classifier.prediction", pred)
# Example prediction:
# Service was bad. (negative)
# First, we clean up the sentence and get a bag of words representation.
instance = bag_of_words(sanitize('Service was bad.'))
# Then, we set up our prediction weights.
# The weights below assume that 'service' and 'was' are not very relevant
# to a positive classification, but the word 'bad' is very non-positive
positive_weights = {'service': 0.2, 'was': 0.0, 'bad': -1}
# The weights below assume that 'service' and 'was' are pretty neutral
neutral_weights = {'service': 0.5, 'was': 0.5, 'bad': 0.1}
# The weights below assume that 'bad' is very negative.
negative_weights = {'service': 0.1, 'was': 0.3, 'bad': 0.9}
# Put all the weights together so we can index into them by their labels
all_weights = {
'positive': positive_weights,
'neutral': neutral_weights,
'negative': negative_weights
}
# Get a list of labels from all of the unique keys (indexes) in the weights
labels = list(set(all_weights.keys()))
# Predict the label of the instance using the weights above and the list of
# possible labels
prediction = predict(instance, all_weights, labels)
print('Predicted label:', prediction[0])
print('Score:', prediction[1])