Correct calculations for when we display/validate/checkpoint

DeepSpeech.ipynb

@@ -1353,8 +1353,13 @@
     "    for epoch in range(start_epoch, training_iters):\n",
     "        params = train_params[:]\n",
     "        \n",
+    "        # Determine if we want to display/validate/checkpoint on this iteration\n",
+    "        is_display_step = display_step > 0 and (epoch + 1) % display_step == 0\n",
+    "        is_validation_step = validation_step > 0 and (epoch + 1) % validation_step == 0\n",
+    "        is_checkpoint_step = (epoch > 0 and (epoch + 1) % checkpoint_step == 0) or epoch == training_iters - 1\n",
+    "        \n",
     "        # Requirements to display a WER report\n",
-    "        if epoch % display_step == 0:\n",
+    "        if is_display_step:\n",
     "            # Reset accuracy\n",
     "            total_accuracy = 0.0\n",
     "            # Create training results tuple\n",
@@ -1382,23 +1387,25 @@
     "            writer.flush()\n",
     "            \n",
     "            # Collect individual sample results for WER report\n",
-    "            if epoch % display_step == 0:\n",
+    "            if is_display_step:\n",
     "                collect_results(train_results, result[2])\n",
     "                # Add batch to total_accuracy\n",
     "                total_accuracy += result[3]\n",
     "        \n",
     "        # Print WER report\n",
-    "        if epoch % display_step == 0:\n",
+    "        if is_display_step:\n",
     "            print \"Epoch:\", '%04d' % (epoch), \"avg_cer=\", \"{:.9f}\".format((total_accuracy / ceil(batches_per_device)))\n",
     "            train_wer = calculate_and_print_wer_report(\"Training\", train_results)\n",
     "            print\n",
+    "        else:\n",
+    "            print \"Epoch: %04d\" % (epoch)\n",
     "            \n",
     "        # Validation step\n",
-    "        if epoch % validation_step == 0:\n",
+    "        if is_validation_step:\n",
     "            dev_wer = run_inference(session, \"Validation\", data_sets.dev.total_batches, feed_dict_validate, results_params)\n",
     "\n",
     "        # Checkpoint the model\n",
-    "        if (epoch % checkpoint_step == 0) or (epoch == training_iters - 1):\n",
+    "        if is_checkpoint_step:\n",
     "            checkpoint_path = os.path.join(checkpoint_dir, 'model.ckpt')\n",
     "            print \"Checkpointing in directory\", \"%s\" % checkpoint_dir\n",
     "            saver.save(session, checkpoint_path, global_step=epoch)\n",