def sample(self, parameters, char_to_ix, sample_limit=50):
vocab_size, n_a = self.n_x, self.n_a
Wx = parameters['Wx']
Wa = parameters['Wa']
Wy = parameters['Wy']
b = parameters['b']
by = parameters['by']
x = np.zeros((vocab_size, 1))
a_prev = np.zeros((n_a, 1))
idx = -1
indices = []
counter = 0
newline_idx = char_to_ix['\n']
while idx != newline_idx and counter != sample_limit:
a = np.tanh(np.dot(Wx, x) + np.dot(Wa, a_prev) + b)
y = softmax(np.dot(Wy, a) + by)
idx = np.random.choice(list(range(vocab_size)), p=y.ravel())
indices.append(idx)
x = np.zeros((vocab_size, 1))
x[idx] = 1
a_prev = a
counter += 1
if counter == 50:
indices.append(newline_idx)
return indices
def step(self, inputs, states):
vP_t = inputs
hP_tm1 = states[0]
_ = states[1:3] # ignore internal dropout/masks
vP, WP_v, WPP_v, v, W_g2 = states[3:8]
vP_mask, = states[8:]
WP_v_Dot = K.dot(vP, WP_v)
WPP_v_Dot = K.dot(K.expand_dims(vP_t, axis=1), WPP_v)
s_t_hat = K.tanh(WPP_v_Dot + WP_v_Dot)
s_t = K.dot(s_t_hat, v)
s_t = K.batch_flatten(s_t)
a_t = softmax(s_t, mask=vP_mask, axis=1)
c_t = K.batch_dot(a_t, vP, axes=[1, 1])
GRU_inputs = K.concatenate([vP_t, c_t])
g = K.dot(GRU_inputs, W_g2)
GRU_inputs = g * GRU_inputs
hP_t, s = super(SelfAttnGRU, self).step(GRU_inputs, states)
return hP_t, s
def predict(self, molecules, return_prob=True):
molecules = deepcopy(molecules)
# Binarize input
if self.binary:
molecules = self.binarize(molecules)
predictions = []
probabilities = []
for molecule in molecules:
P = [0] * self.num_classes
for c in self.classes:
# add log prior
P[c] = np.log(self.P_c[c])
for atom in molecule:
if atom >= 0:
# calculate log likelihood. Molecules has pad as index 0, so shift it one
P[c] += np.log(self.p_w_given_c[int(atom), c])
# normalize the posterior distribution, using softmax
P = softmax(P)
predictions.append(np.argmax(P))
probabilities.append(P)
if return_prob:
return predictions, probabilities
return predictions
def run_inference(inf_file):
# Preprocessing of the image happens here
img = load_image(inf_file)
print("Image Loaded")
img = make_square(img)
img = augment(prep, img)
print("Transformations done")
img = img.transpose(-1, 0, 1).astype(np.float32)
img = img.reshape(-1, CHANNELS, IMG_WIDTH, IMG_HEIGHT)
# Inferencing starts here
sess = onnxruntime.InferenceSession("./best_acc.onnx")
print("The model expects input shape: ", sess.get_inputs()[0].shape)
print("The shape of the Image is: ", img.shape)
input_name = sess.get_inputs()[0].name
result = sess.run(None, {input_name: img})
prob_array = result[0][0]
print("Prob Array ", prob_array)
prob = max(softmax(result[0][0]))
print("Prob ", prob)
species = tf.argmax(prob_array.ravel()[:10]).numpy()
print("Class Label ", species)
print("Spec ", CLASS_MAP[species][1])
string_label = CLASS_MAP[species][1].split(" ")
return (string_label[0], string_label[1], str(prob), color_code(prob))
def feedForward(self, inputRow):
hiddenNet = (self.hiddenLayerWeights @ inputRow) + self.hiddenLayerBiases
hiddenActivation = self.activationFun(hiddenNet)
outputNet = (self.outputLayerWeights @ hiddenActivation) + self.outputLayerBiases
outputActivation = helpers.softmax(outputNet)
self.hiddenLayerNetValues = hiddenNet
self.hiddenLayerActivations.append(hiddenActivation)
self.outputLayerNetValues = outputNet
self.outputLayerActivations.append(outputActivation)
if self.useNag:
hiddenNagNet = ((self.hiddenLayerWeights - self.hiddenMomentum) @ inputRow) + (self.hiddenLayerBiases - self.hiddenBiasMomentum)
outputNagNet = ((self.outputLayerWeights - self.outputMomentum) @ hiddenActivation) + (self.outputLayerBiases - self.outputBiasMomentum)
outputNagActivation = helpers.softmax(outputNagNet)
self.hiddenNagNetValues = hiddenNagNet
self.outputNagActivation = outputNagActivation
return outputActivation
def forward(self, xt, a_prev, parameters):
Wx = parameters['Wx']
Wa = parameters['Wa']
Wy = parameters['Wy']
b = parameters['b']
by = parameters['by']
a_next = np.tanh(np.dot(Wx, xt) + np.dot(Wa, a_prev) + b)
yt = softmax(np.dot(Wy, a_next) + by)
cache = {'a_prev': a_prev, 'a_next': a_next, 'xt': xt, 'yt': yt}
return a_next, yt
def call(self, inputs, mask=None):
assert (isinstance(inputs, list) and len(inputs) == 5)
uQ, WQ_u, WQ_v, v, VQ_r = inputs
uQ_mask = mask[0] if mask is not None else None
ones = K.ones_like(K.sum(uQ, axis=1, keepdims=True)) # (B, 1, 2H)
s_hat = K.dot(uQ, WQ_u)
s_hat += K.dot(ones, K.dot(WQ_v, VQ_r))
s_hat = K.tanh(s_hat)
s = K.dot(s_hat, v)
s = K.batch_flatten(s)
a = softmax(s, mask=uQ_mask, axis=1)
rQ = K.batch_dot(uQ, a, axes=[1, 1])
return rQ
def forward_pass(self, inp):
for ix in range(len(inp)):
if inp[ix] == 0:
self.activation[0][ix] = 1
else:
self.activation[0][ix] = 0
self.X = self.activation
self.L1 = np.tanh(np.dot(self.X, self.W1))
self.L2 = np.tanh(np.dot(self.L1, self.W2))
self.Y = softmax(np.dot(self.L2, self.W3))
self.y_out = []
for ix in range(self.Y.shape[1]):
self.y_out.append(self.Y[0][ix] * self.activation[0][ix])
self.OUT = np.argmax(self.y_out)
return self.OUT
def step(self, inputs, states):
# input
ha_tm1 = states[0] # (B, 2H)
_ = states[1:3] # ignore internal dropout/masks
hP, WP_h, Wa_h, v = states[3:7] # (B, P, 2H)
hP_mask, = states[7:8]
WP_h_Dot = K.dot(hP, WP_h) # (B, P, H)
Wa_h_Dot = K.dot(K.expand_dims(ha_tm1, axis=1), Wa_h) # (B, 1, H)
s_t_hat = K.tanh(WP_h_Dot + Wa_h_Dot) # (B, P, H)
s_t = K.dot(s_t_hat, v) # (B, P, 1)
s_t = K.batch_flatten(s_t) # (B, P)
a_t = softmax(s_t, mask=hP_mask, axis=1) # (B, P)
c_t = K.batch_dot(hP, a_t, axes=[1, 1]) # (B, 2H)
GRU_inputs = c_t
ha_t, (ha_t_, ) = super(PointerGRU, self).step(GRU_inputs, states)
return a_t, [ha_t]
def run_inference(inf_file):
# Preprocessing of the image happens here
img = load_image(inf_file)
originalimg = img
#Cropping line is below, SHOULD NOT BE INCLUDED IN THIS VERSION
#useless, img, status=cropImage(impath, 'm', labelsfile, 21, 0.08)
print("Image Loaded")
img = make_square(img)
img = augment(prep, img)
print("Transformations done")
img = img.transpose(-1, 0, 1).astype(np.float32)
img = img.reshape(-1, CHANNELS, IMG_WIDTH, IMG_HEIGHT)
# Inferencing starts here
sess = onnxruntime.InferenceSession("./best_acc.onnx")
print("The model expects input shape: ", sess.get_inputs()[0].shape)
print("The shape of the Image is: ", img.shape)
input_name = sess.get_inputs()[0].name
result = sess.run(None, {input_name: img})
prob_array = result[0][0]
print("Prob Array ", prob_array)
prob = max(softmax(result[0][0]))
print("Prob ", prob)
species = tf.argmax(prob_array.ravel()[:20]).numpy()
print("Class Label ", species)
print("Spec ", CLASS_MAP[species][1])
string_label = CLASS_MAP[species][1].split(" ")
#fullfilename = mosquitoid + "_" + picnum + "_" + string_label[0] + "_" + string_label[1]
#fullbucket = 'photostakenduringpilotstudy'
#s3_file = 'PilotStudy'
#file = cv2.imwrite(fullfilename, originalimg)
#upload_to_aws(file, fullbucket , s3_file)
##upload_file(file, fullbucket)
return (string_label[0], string_label[1], str(prob), color_code(prob))
def step(self, inputs, states):
uP_t = inputs
vP_tm1 = states[0]
_ = states[1:3] # ignore internal dropout/masks
uQ, WQ_u, WP_v, WP_u, v, W_g1 = states[3:9]
uQ_mask, = states[9:10]
WQ_u_Dot = K.dot(uQ, WQ_u) #WQ_u
WP_v_Dot = K.dot(K.expand_dims(vP_tm1, axis=1), WP_v) #WP_v
WP_u_Dot = K.dot(K.expand_dims(uP_t, axis=1), WP_u) # WP_u
s_t_hat = K.tanh(WQ_u_Dot + WP_v_Dot + WP_u_Dot)
s_t = K.dot(s_t_hat, v) # v
s_t = K.batch_flatten(s_t)
a_t = softmax(s_t, mask=uQ_mask, axis=1)
c_t = K.batch_dot(a_t, uQ, axes=[1, 1])
GRU_inputs = K.concatenate([uP_t, c_t])
g = K.sigmoid(K.dot(GRU_inputs, W_g1)) # W_g1
GRU_inputs = g * GRU_inputs
vP_t, s = super(QuestionAttnGRU, self).step(GRU_inputs, states)
return vP_t, s
