GC seems to pass for LSTM o.o

xperiments/xp_dense.py

@@ -7,10 +7,10 @@
 inshape, outshape = X.shape[1:], Y.shape[1:]
 
 network = BackpropNetwork(input_shape=inshape, layerstack=[
-    DenseLayer(64, activation="tanh", trainable=1),
-    DenseLayer(32, activation="tanh", trainable=1),
-    DenseLayer(outshape, activation="sigmoid", trainable=1)
-], cost="bxent", optimizer="sgd")
+    DenseLayer(32, activation="sigmoid", trainable=1),
+    DenseLayer(32, activation="sigmoid", trainable=1),
+    DenseLayer(outshape, activation="linear", trainable=1)
+], cost="mse", optimizer="sgd")
 network.fit(X[5:], Y[5:], epochs=1, batch_size=len(X)-5, verbose=0)
 
 gcsuite = GradientCheck(network, epsilon=1e-7)

xperiments/xp_lstm.py

@@ -4,18 +4,17 @@
 from brainforge.layers import LSTM, DenseLayer
 from brainforge.gradientcheck import GradientCheck
 
-np.random.seed(1337)
+# np.random.seed(1337)
 
-DSHAPE = 10, 1, 15
-OUTSHP = 10, 15
+DSHAPE = 20, 10, 1
+OUTSHP = 20, 1
 X = np.random.randn(*DSHAPE)
 Y = np.random.randn(*OUTSHP)
 
 net = BackpropNetwork(input_shape=DSHAPE[1:], layerstack=[
-    LSTM(32, activation="tanh"),
-    DenseLayer(10, activation="tanh", trainable=False),
-    DenseLayer(OUTSHP[1:], activation="linear", trainable=False)
+    LSTM(16, activation="tanh"),
+    DenseLayer(OUTSHP[1:], activation="linear", trainable=0)
 ], cost="mse", optimizer="sgd")
 
 net.fit(X, Y, epochs=1, verbose=0)
-GradientCheck(net, epsilon=1e-6, display=True).run(X, Y, throw=True)
+GradientCheck(net, display=True).run(X, Y, throw=True)

xperiments/xp_pggymin.py

@@ -5,7 +5,7 @@
 
 from matplotlib import pyplot
 
-from brainforge import BackpropNetwork
+from brainforge.learner import BackpropNetwork
 from brainforge.layers import DenseLayer
 from brainforge.optimization import Momentum
 from brainforge.reinforcement import PG, AgentConfig