- some early stopping and change of last pooling layer

few_shot_tests.py

@@ -11,20 +11,20 @@
 from mini_imagenet_dataloader import MiniImageNetDataLoader
 from tensorflow.keras.models import Model
 from tensorflow.keras.layers import Activation, Dense, Input, Flatten, Conv2D, Lambda, TimeDistributed, MaxPooling2D, Layer
-from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard
+from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard, Callback
 import tensorflow.keras.backend as K
 import tensorflow.keras
 from tensorflow.keras import optimizers as op
 import tensorflow as tf
 import argparse
 import numpy as np
 import random, imageio
-
+import signal
 
 # To support editing of command line parameters use the fork https://github.com/dsmic/Gooey
 # uncomment the following 2 lines for standard command line handling without gui
 from gooey import Gooey
-@Gooey(load_cmd_args=True)
+@Gooey(load_cmd_args=True, ignore_command='--no-gui', force_command='--gui')
 
 def parser():
     global args
@@ -54,10 +54,20 @@ def parser():
     parser.add_argument('--cathegories', dest='cathegories',  type=int, default=5)
     parser.add_argument('--only_one_samplefolder', dest='only_one_samplefolder', action='store_true')
     parser.add_argument('--load_subnet', dest='load_subnet', action='store_true')
+    parser.add_argument('--EarlyStop', dest='EarlyStop',  type=str, default='EarlyStop')
     args = parser.parse_args()
 
 parser()
 
+flag=0
+def CTRL_C(sig, frame):
+    global flag
+    print(flag)
+    #import code; code.interact()
+    import pdb; pdb.set_trace()
+
+signal.signal(signal.SIGINT, CTRL_C)
+
 # uncomment the following to disable CuDNN support
 #import os
 #os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
@@ -358,7 +368,7 @@ def get_config(self):
 conv4 = TimeDistributed(Conv2D(args.hidden_size, 3, padding='same', activation = 'relu', name = 'conv_4'))(pool3)
 pool4 = TimeDistributed(MaxPooling2D(pool_size = 2, name = 'pool_4'))(conv4)
 conv5 = TimeDistributed(Conv2D(args.hidden_size, 3, padding='same', activation = 'relu', name = 'conv_5'))(pool4)
-pool5 = TimeDistributed(MaxPooling2D(pool_size = 2, name = 'pool_5'))(conv5)
+pool5 = TimeDistributed(MaxPooling2D(pool_size = 5, name = 'pool_5'))(conv5)
 
 flat = TimeDistributed(Flatten())(pool5)
 #x = TimeDistributed(Dense(100, activation = 'relu'))(flat)
@@ -483,10 +493,18 @@ def all_layers(model):
 
 #print('vor fitting', lambda_model_layers[17].get_weights()[0])
 
+import os
+class TerminateKey(Callback):
+    def on_batch_end(self, batch, logs=None):
+        if os.path.exists(args.EarlyStop):
+            self.model.stop_training = True
+
+terminate_on_key = TerminateKey()
+
 checkpointer = ModelCheckpoint(filepath='checkpoints/model-{epoch:02d}.hdf5', verbose=1)
 tensorboard = TensorBoard(log_dir = args.tensorboard_logdir)
 lambda_model.fit_generator(keras_gen_train.generate_add_samples(), train_epoch_size, args.epochs, 
-                           validation_data=keras_gen_train.generate_add_samples('test'), validation_steps=test_epoch_size, callbacks = [tensorboard], workers = 0) 
+                           validation_data=keras_gen_train.generate_add_samples('test'), validation_steps=test_epoch_size, callbacks = [tensorboard, terminate_on_key], workers = 0)
 #workers = 0 is a work around to correct the number of calls to the validation_data generator
 
 #test_lambda = lambda_model([K.expand_dims(K.variable(base_train_img[0:0+1]),axis=0),K.expand_dims(K.variable(base_train_img), axis=0), K.expand_dims(K.variable(base_train_label), axis=0)])
@@ -565,6 +583,11 @@ def print_FindModels(model):
 lambda_model.save_weights(args.final_name+'-weights.hdf5')
 lambda_model.layers[2].save_weights(args.final_name + '-weights.hdf5' + '_subnet.hdf5')
 
+
+if os.path.exists(args.EarlyStop) and os.path.getsize(args.EarlyStop)==0:
+    os.remove(args.EarlyStop)
+    print('removed',args.EarlyStop)
+
 # tools for debugging
 def get_weight_grad(model, inputs, outputs):
     """ Gets gradient of model for given inputs and outputs for all weights"""