fixed error for other WIN sizes

JayRnn.py

@@ -24,16 +24,16 @@ from rbm.train import train
 from rbm.status import Status
 
 # ── Hyper-parameters ──────────────────────────────────────────────────────────
-TEXT    = "  HALLO SUPER JENS!" # the character sequence to learn (matches diagram example)
+TEXT    = "  HALLO SUPER JENS UND SUPER MAUSI!" # the character sequence to learn (matches diagram example)
-WIN    = 3         # sliding-window width  (= N in the diagram)
+WIN    = 5         # sliding-window width  (= N in the diagram)
 STRIDE = 1     	   # sliding-window step size
 UNITS  = 1
-H_SIZE = 128         # hidden units per RBM cell
+H_SIZE = 180         # hidden units per RBM cell
-NUM_EPOCHS = 1000
+NUM_EPOCHS = 100
-NUM_ITERATIONS = 1
+NUM_ITERATIONS = 10
 
 TRAIN_PARAMS = TrainingParams(
-	learning_rate    = 0.25,
+	learning_rate    = 0.04,
 	momentum         = 0.5,
 	num_epochs       = NUM_EPOCHS,
 	do_rao_blackwell = True,
@@ -125,14 +125,20 @@ if __name__ == "__main__":
 		model.save()
 
 	# test the model
-	seed = str2vec('HA^')
+	seed_str = 'HA^'
-	v_forward = seed.flatten()
+	seed_padded = ' '*(WIN-len(seed_str)) + seed_str
+	v_forward = str2vec(seed_padded).flatten()
 	context = c_train[0]
+	text_predict = ''
 	for i in range(len(TEXT)+5):
 		v_mat = v_forward.reshape([WIN, vocab_size()])
 		print(f"Forward {i:02d}: {vec2str(v_mat, axis=1)}")
 		v_predict, context = model.forward_step(v_mat, context)
-		v_predict_clamp = clamp(v_predict.reshape([WIN, vocab_size()]), axis=1)
+		v_predict_mat = v_predict.reshape([WIN, vocab_size()])
-		print(f"Predict {i:02d}: {vec2str(v_predict_clamp, axis=1)}")
+		v_predict_clamp = clamp(v_predict_mat, axis=1)
+		v_predict_str = vec2str(v_predict_clamp, axis=1)
+		text_predict += v_predict_str[WIN-1]
+		print(f"Predict {i:02d}: {v_predict_str}")
 		v_forward = shift_left(v_predict_clamp.reshape([1, WIN*vocab_size()]), vocab_size())
 
+	print(seed_padded[0:WIN-1] + text_predict)