Minor fix logging.

deepmedic/logging/utils.py

@@ -77,11 +77,21 @@ def getMeanPerColOf2dListExclNA(list2d, notApplicPattern) :
     return listWithMeanPerColumn
 
 
-
 def datetime_now_str() :
     #datetime returns in the format: YYYY-MM-DD HH:MM:SS.millis but ':' is not supported for Windows' naming convention.
     dateTimeNowStr = str(datetime.datetime.now())
     dateTimeNowStr = dateTimeNowStr.replace(':','.')
     dateTimeNowStr = dateTimeNowStr.replace(' ','.')
     return dateTimeNowStr
 
+# TODO: Merge the two below
+def print_progress_step_tr_val(log, n_batches, n_batches_done, batch_size, prefix_str):
+    progress_step = max(1, n_batches // 5)
+    if n_batches_done == 0 or (n_batches_done % progress_step) == 0 or (n_batches_done == n_batches):
+        log.print3(prefix_str + " Processed " + str(n_batches_done) + "/" + str(n_batches) + " batches for this subepoch...")
+
+def print_progress_step_test(log, n_batches, n_batches_done, batch_size, n_tiles_for_subj):
+    progress_step = max(1, n_batches // 5)
+    if n_batches_done == 0 or (n_batches_done % progress_step) == 0 or (n_batches_done == n_batches):
+        log.print3("Processed " + str(n_batches_done * batch_size) + "/" + str(n_tiles_for_subj) + " segments.")
+

deepmedic/routines/testing.py

@@ -18,9 +18,8 @@
 from deepmedic.dataManagement.io import savePredImgToNiiWithOriginalHdr, saveFmImgToNiiWithOriginalHdr, \
     save4DImgWithAllFmsToNiiWithOriginalHdr
 from deepmedic.dataManagement.preprocessing import unpad_3d_img
-
 from deepmedic.neuralnet.pathwayTypes import PathwayTypes as pt
-from deepmedic.logging.utils import strListFl4fNA, getMeanPerColOf2dListExclNA
+from deepmedic.logging.utils import strListFl4fNA, getMeanPerColOf2dListExclNA, print_progress_step_test
 
 
 
@@ -202,13 +201,6 @@ def stitch_predicted_to_fms(array_fms_to_save, idx_next_tile_in_fm_vols,
     return idx_next_tile_in_fm_vols, array_fms_to_save
 
 
-def print_progress_step(log, n_batches, batch_i, batch_size, n_tiles_for_subj):
-    progress_step = max(1, n_batches // 5)
-
-    if batch_i == 0 or ((batch_i + 1) % progress_step) == 0 or (batch_i + 1) == n_batches:
-        log.print3("Processed " + str(batch_i * batch_size) + "/" + str(n_tiles_for_subj) + " segments.")
-
-
 def prepare_feeds_dict(feeds, channs_of_tiles_per_path):
     # TODO: Can we rename the input feeds so that they are easier to deal with?
     feeds_dict = {feeds['x']: np.asarray(channs_of_tiles_per_path[0], dtype='float32')}
@@ -258,9 +250,9 @@ def predict_whole_volume_by_tiling(log, sessionTf, cnn3d,
     idx_next_tile_in_fm_vols = 0
     n_batches = n_tiles_for_subj // batchsize
     t_fwd_pass_subj = 0 # time it took for forward pass over all tiles of subject.
+    print_progress_step_test(log, n_batches, 0, batchsize, n_tiles_for_subj)    
     for batch_i in range(n_batches):
 
-        print_progress_step(log, n_batches, batch_i, batchsize, n_tiles_for_subj)
         # Extract data for the segments of this batch.
         # ( I could modularize extractDataOfASegmentFromImagesUsingSampledSliceCoords()
         # of training and use it here as well. )
@@ -307,6 +299,7 @@ def predict_whole_volume_by_tiling(log, sessionTf, cnn3d,
                                                           outp_pred_dims,
                                                           cnn3d.pathways,
                                                           idxs_fms_to_save)
+        print_progress_step_test(log, n_batches, batch_i + 1, batchsize, n_tiles_for_subj)
         # Done with batch
 
     log.print3("TIMING: Segmentation of subject: [Forward Pass:] {0:.2f}".format(t_fwd_pass_subj) + " secs.")

deepmedic/routines/training.py

@@ -20,7 +20,7 @@
 from deepmedic.dataManagement.sampling import get_samples_for_subepoch
 from deepmedic.routines.testing import inference_on_whole_volumes
 
-from deepmedic.logging.utils import datetime_now_str
+from deepmedic.logging.utils import datetime_now_str, print_progress_step_tr_val
 
 
 def process_in_batches(log,
@@ -38,16 +38,12 @@ def process_in_batches(log,
     # Each row of array below holds number of:
     #     Real Positives, Real Neg, True Predicted Pos, True Predicted Neg in subepoch, in this order.
     arr_RpRnTpTn_per_class_in_subep = np.zeros([cnn3d.num_classes, 4], dtype="int32")
-
-    print_progress_step = max(1, n_batches // 5)
-
+    
+    prefix_progress_str = '[TRAINING]' if train_or_val == 'train' else '[VALIDATION]'
+    print_progress_step_tr_val(log, n_batches, 0, batchsize, prefix_progress_str)
     for batch_i in range(n_batches):
 
         if train_or_val == "train":
-            if batch_i == 0 or ((batch_i + 1) % print_progress_step) == 0 or (batch_i + 1) == n_batches:
-                log.print3("[TRAINING] Trained on " + str(batch_i) + "/" + str(n_batches) +\
-                           " batches for this subepoch...")
-
             ops_to_fetch = cnn3d.get_main_ops('train')
             list_of_ops = [ops_to_fetch['cost']] + ops_to_fetch['list_rp_rn_tp_tn'] +\
                             [ops_to_fetch['updates_grouped_op']]
@@ -68,12 +64,8 @@ def process_in_batches(log,
 
             cost_this_batch = results_of_run[0]
             list_RpRnPpPn_per_class = results_of_run[1:-1]  # [-1] is from updates_grouped_op, returns nothing
-
+            
         else:  # validation
-            if batch_i == 0 or ((batch_i + 1) % print_progress_step) == 0 or (batch_i + 1) == n_batches:
-                log.print3("[VALIDATION] Validated on " +
-                           str(batch_i) + "/" + str(n_batches) + " batches for this subepoch...")
-
             ops_to_fetch = cnn3d.get_main_ops('val')
             list_of_ops = ops_to_fetch['list_rp_rn_tp_tn']
 
@@ -91,7 +83,7 @@ def process_in_batches(log,
 
             cost_this_batch = 999  # placeholder in case of validation.
             list_RpRnPpPn_per_class = results_of_run
-
+        
         # list_RpRnPpPn_per_class holds Real Pos, Real Neg, True Pred Pos, True Pred Neg ...
         # ... for all classes, in this order, flattened. First RpRnTpTn are for 'WHOLE' class.
         arr_RpRnTpTn_per_class = np.asarray(list_RpRnPpPn_per_class, dtype="int32")
@@ -101,6 +93,8 @@ def process_in_batches(log,
         costs_of_batches.append(cost_this_batch)  # only really used in training.
         arr_RpRnTpTn_per_class_in_subep += arr_RpRnTpTn_per_class
 
+        print_progress_step_tr_val(log, n_batches, batch_i + 1, batchsize, prefix_progress_str)
+
     # ======== Calculate and Report accuracy over subepoch
     # In case of validation, mean_cost_subep is just a placeholder.
     # Cause this does not get calculated and reported in this case.