reshape(())

chapter_optimization/adadelta-scratch.md

@@ -109,7 +109,7 @@ def optimize(batch_size, rho, num_epochs, log_interval):
                 y_vals.append(squared_loss(net(X, w, b), y).mean().asnumpy())
         print('epoch %d, loss %.4e' % (epoch, y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', w.reshape(1, -1).asnumpy(), 'b:', b.asscalar(), '\n')
+    print('w:', w.reshape((1, -1)).asnumpy(), 'b:', b.asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     utils.set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)

chapter_optimization/adagrad-scratch.md

@@ -126,7 +126,7 @@ def optimize(batch_size, lr, num_epochs, log_interval):
         print('epoch %d, learning rate %f, loss %.4e' % (epoch, lr,
                                                          y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', w.reshape(1, -1).asnumpy(), 'b:', b.asscalar(), '\n')
+    print('w:', w.reshape((1, -1)).asnumpy(), 'b:', b.asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     utils.set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)

chapter_optimization/adam-scratch.md

@@ -146,7 +146,7 @@ def optimize(batch_size, lr, num_epochs, log_interval):
         print('epoch %d, learning rate %f, loss %.4e' % (epoch, lr,
                                                          y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', w.reshape(1, -1).asnumpy(), 'b:', b.asscalar(), '\n')
+    print('w:', w.reshape((1, -1)).asnumpy(), 'b:', b.asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     utils.set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)

chapter_optimization/gd-sgd-scratch.md

@@ -194,7 +194,7 @@ def optimize(batch_size, lr, num_epochs, log_interval, decay_epoch):
         print('epoch %d, learning rate %f, loss %.4e' % (epoch, lr,
                                                          y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', w.reshape(1, -1).asnumpy(), 'b:', b.asscalar(), '\n')
+    print('w:', w.reshape((1, -1)).asnumpy(), 'b:', b.asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     utils.set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)

chapter_optimization/momentum-scratch.md

@@ -122,7 +122,7 @@ def optimize(batch_size, lr, mom, num_epochs, log_interval):
         print('epoch %d, learning rate %f, loss %.4e' % (epoch, lr,
                                                          y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', w.reshape(1, -1).asnumpy(), 'b:', b.asscalar(), '\n')
+    print('w:', w.reshape((1, -1)).asnumpy(), 'b:', b.asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     utils.set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)

chapter_optimization/rmsprop-scratch.md

@@ -108,7 +108,7 @@ def optimize(batch_size, lr, gamma, num_epochs, log_interval):
         print('epoch %d, learning rate %f, loss %.4e' % (epoch, lr,
                                                          y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', w.reshape(1, -1).asnumpy(), 'b:', b.asscalar(), '\n')
+    print('w:', w.reshape((1, -1)).asnumpy(), 'b:', b.asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     utils.set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)

utils.py

@@ -401,8 +401,8 @@ def optimize(batch_size, trainer, num_epochs, decay_epoch, log_interval, X, y,
         else:
             print("epoch %d, loss %.4e" % (epoch, y_vals[-1]))
     # 为了便于打印，改变输出形状并转化成numpy数组。
-    print('w:', np.reshape(net[0].weight.data().asnumpy(), (1, -1)),
-          'b:', net[0].bias.data().asnumpy()[0], '\n')
+    print('w:', net[0].weight.data().reshape((1, -1)).asnumpy(),
+          'b:', net[0].bias.data().asscalar(), '\n')
     x_vals = np.linspace(0, num_epochs, len(y_vals), endpoint=True)
     set_fig_size(mpl)
     plt.semilogy(x_vals, y_vals)