From cc285186993a66e9886957b2b5c106ec56c9949e Mon Sep 17 00:00:00 2001
From: Jean-Marc Valin <jmvalin@jmvalin.ca>
Date: Sat, 19 Dec 2020 19:25:59 -0500
Subject: [PATCH] wip 8x4 sparseness

---
 dnn/training_tf2/dump_lpcnet.py  | 10 ++++++----
 dnn/training_tf2/lpcnet.py       |  8 +++++---
 dnn/training_tf2/train_lpcnet.py |  4 ++--
 3 files changed, 13 insertions(+), 9 deletions(-)

diff --git a/dnn/training_tf2/dump_lpcnet.py b/dnn/training_tf2/dump_lpcnet.py
index c42d2107..ce64c614 100755
--- a/dnn/training_tf2/dump_lpcnet.py
+++ b/dnn/training_tf2/dump_lpcnet.py
@@ -66,15 +66,17 @@ def printSparseVector(f, A, name):
     A[:,2*N:] = A[:,2*N:] - np.diag(np.diag(A[:,2*N:]))
     printVector(f, diag, name + '_diag')
     idx = np.zeros((0,), dtype='int')
-    for i in range(3*N//16):
+    for i in range(3*N//8):
         pos = idx.shape[0]
         idx = np.append(idx, -1)
         nb_nonzero = 0
-        for j in range(N):
-            if np.sum(np.abs(A[j, i*16:(i+1)*16])) > 1e-10:
+        for j in range(N//4):
+            block = A[j*4:(j+1)*4, i*8:(i+1)*8]
+            if np.sum(np.abs(block)) > 1e-10:
                 nb_nonzero = nb_nonzero + 1
                 idx = np.append(idx, j)
-                W = np.concatenate([W, A[j, i*16:(i+1)*16]])
+                vblock = block.transpose((1,0)).reshape((-1,))
+                W = np.concatenate([W, vblock])
         idx[pos] = nb_nonzero
     printVector(f, W, name)
     #idx = np.tile(np.concatenate([np.array([N]), np.arange(N)]), 3*N//16)
diff --git a/dnn/training_tf2/lpcnet.py b/dnn/training_tf2/lpcnet.py
index 546e6d22..2e61f1d9 100644
--- a/dnn/training_tf2/lpcnet.py
+++ b/dnn/training_tf2/lpcnet.py
@@ -74,12 +74,14 @@ class Sparsify(Callback):
                 A = p[:, k*N:(k+1)*N]
                 A = A - np.diag(np.diag(A))
                 #A = np.transpose(A, (1, 0))
-                L=np.reshape(A, (N, N//16, 16))
+                L=np.reshape(A, (N//4, 4, N//8, 8))
                 S=np.sum(L*L, axis=-1)
+                S=np.sum(S, axis=1)
                 SS=np.sort(np.reshape(S, (-1,)))
-                thresh = SS[round(N*N//16*(1-density))]
+                thresh = SS[round(N*N//32*(1-density))]
                 mask = (S>=thresh).astype('float32');
-                mask = np.repeat(mask, 16, axis=1)
+                mask = np.repeat(mask, 4, axis=0)
+                mask = np.repeat(mask, 8, axis=1)
                 mask = np.minimum(1, mask + np.diag(np.ones((N,))))
                 #mask = np.transpose(mask, (1, 0))
                 p[:, k*N:(k+1)*N] = p[:, k*N:(k+1)*N]*mask
diff --git a/dnn/training_tf2/train_lpcnet.py b/dnn/training_tf2/train_lpcnet.py
index f6eddccb..7dd31907 100755
--- a/dnn/training_tf2/train_lpcnet.py
+++ b/dnn/training_tf2/train_lpcnet.py
@@ -102,7 +102,7 @@ del pred
 del in_exc
 
 # dump models to disk as we go
-checkpoint = ModelCheckpoint('lpcnet32c_384_10_G16_{epoch:02d}.h5')
+checkpoint = ModelCheckpoint('lpcnet32v_384_10_G16_{epoch:02d}.h5')
 
 #Set this to True to adapt an existing model (e.g. on new data)
 adaptation = False
@@ -120,5 +120,5 @@ else:
     decay = 5e-5
 
 model.compile(optimizer=Adam(lr, decay=decay, beta_2=0.99), loss='sparse_categorical_crossentropy')
-model.save_weights('lpcnet32c_384_10_G16_00.h5');
+model.save_weights('lpcnet32v_384_10_G16_00.h5');
 model.fit([in_data, features, periods], out_exc, batch_size=batch_size, epochs=nb_epochs, validation_split=0.0, callbacks=[checkpoint, sparsify])