def loadPredictions(self, path_model):
print("============== loading model =============")
model = load_model(path_model, compile=False)
print("Model", model)
print("Loading in data: ", self.path_test + 'patches_in.npy')
test_in = np.load(self.path_test + 'patches_in.npy', mmap_mode='r')
# test_label=np.load(self.path_test+'patches_label.npy')
test_label = np.load(self.path_test + 'patches_label.npy')[:,
-1] # may18
# add doty
# if dataset=='lm':
ds = LEM()
dataSource = SARSource()
ds.addDataSource(dataSource)
dotys, dotys_sin_cos = ds.getDayOfTheYear()
def addDoty(input_):
input_ = [input_, dotys_sin_cos]
return input_
dotys_sin_cos = np.expand_dims(dotys_sin_cos,
axis=0) # add batch dimension
dotys_sin_cos = np.repeat(dotys_sin_cos, test_in.shape[0], axis=0)
#test_in = addDoty(test_in)
# Here do N to 1 prediction for last timestep at first...
test_predictions = model.predict(test_in)
print(test_in[0].shape, test_label.shape, test_predictions.shape)
print("Test predictions dtype", test_predictions.dtype)
#pdb.set_trace()
del test_in
return test_predictions, test_label
full_label_test[full_label_test == 255] = class_n
print(full_ims_test.shape)
print(full_label_test.shape)
# add doty
mim = MIMFixed()
data = {'labeled_dates': 12}
data['labeled_dates'] = 12
if dataset == 'lm':
ds = LEM()
elif dataset == 'l2':
ds = LEM2()
dataSource = SARSource()
ds.addDataSource(dataSource)
time_delta = ds.getTimeDelta(delta=True, format='days')
ds.setDotyFlag(True)
dotys, dotys_sin_cos = ds.getDayOfTheYear()
ds.dotyReplicateSamples(sample_n=1)
# Reconstruct the image
print("Reconstructing the labes and predictions...")
patch_size = 32
add_padding_flag = False
if add_padding_flag == True:
full_ims_test, stride, step_row, step_col, overlap = seq_add_padding(
full_ims_test, patch_size, 0)
def loadPredictions(self, path_model):
print("============== loading model =============")
model = load_model(path_model, compile=False)
print("Model", model)
print("Loading in data: ", self.path_test + 'patches_in.npy')
batch = {}
batch['in'] = np.load(self.path_test + 'patches_in.npy',
mmap_mode='r') # len is 21
# test_label=np.load(self.path_test+'patches_label.npy')
self.labeled_dates = 12
batch['label'] = np.load(self.path_test + 'patches_label.npy')
deb.prints(batch['label'].shape)
#pdb.set_trace()
batch['label'] = batch['label'][:, -self.labeled_dates:] # may18
deb.prints(batch['in'].shape)
deb.prints(batch['label'].shape)
#pdb.set_trace()
self.mim = MIMVarLabel_PaddedSeq()
# self.mim = MIMFixed()
data = {'labeled_dates': 12}
data['labeled_dates'] = 12
#batch = {'in': test_in, 'label': test_label}
# add doty
if self.dataset == 'lm':
ds = LEM()
elif self.dataset == 'l2':
ds = LEM2()
dataSource = SARSource()
ds.addDataSource(dataSource)
time_delta = ds.getTimeDelta(delta=True, format='days')
ds.setDotyFlag(True)
dotys, dotys_sin_cos = ds.getDayOfTheYear()
ds.dotyReplicateSamples(sample_n=batch['label'].shape[0])
prediction_dtype = np.float16
model_t_len = 12
batch['shape'] = (batch['in'].shape[0],
model_t_len) + batch['in'].shape[2:]
# get model class n
model_shape = model.layers[-1].output_shape
model_class_n = model_shape[-1]
deb.prints(model_shape)
deb.prints(model_class_n)
test_predictions = np.zeros_like(batch['label'][..., :-1],
dtype=prediction_dtype)
# test_predictions = np.zeros((batch['label'].shape[:-1])+(model_class_n,), dtype = prediction_dtype)
deb.prints(test_predictions.shape)
#pdb.set_trace()
for t_step in range(data['labeled_dates']): # 0 to 11
###batch_val_label = batch['label'][:, t_step]
#data.patches['test']['label'] = data.patches['test']['label'][:, label_id]
##deb.prints(batch_val_label.shape)
##deb.prints(t_step-data['labeled_dates'])
#pdb.set_trace()
input_ = self.mim.batchTrainPreprocess(
batch, ds, label_date_id=t_step -
data['labeled_dates']) # tstep is -12 to -1
deb.prints(input_[0].shape)
#pdb.set_trace()
#deb.prints(data.patches['test']['label'].shape)
test_predictions[:, t_step] = (
model.predict(input_)).astype(prediction_dtype)
#pdb.set_trace()
print(" shapes", test_predictions, batch['label'])
print(
"batch['in'][0].shape, batch['label'].shape, test_predictions.shape",
batch['in'][0].shape, batch['label'].shape, test_predictions.shape)
print("Test predictions dtype", test_predictions.dtype)
deb.prints(
np.unique(test_predictions.argmax(axis=-1), return_counts=True))
deb.prints(
np.unique(batch['label'].argmax(axis=-1), return_counts=True))
#test_predictions = test_predictions.argmax(axis=-1)
#batch['label'] = batch['label'].argmax(axis=-1)
print("uniques",
np.unique(test_predictions.argmax(axis=-1), return_counts=True),
np.unique(batch['label'].argmax(axis=-1), return_counts=True))
#test_predictions = self.newLabel2labelTranslate(test_predictions, 'new_labels2labels_lm_20171209_S1.pkl')
#batch['label'] = self.newLabel2labelTranslate(batch['label'], 'new_labels2labels_l2_20191223_S1.pkl')
print("uniques",
np.unique(test_predictions.argmax(axis=-1), return_counts=True),
np.unique(batch['label'].argmax(axis=-1), return_counts=True))
#pdb.set_trace()
del batch['in']
return test_predictions, batch['label']