generators should be indexed from 1

.gitignore

@@ -99,3 +99,8 @@ ENV/
 
 # mypy
 .mypy_cache/
+
+*.gz
+*.t7
+
+out

began.py

@@ -126,7 +126,8 @@ def began_disc(x, name, **kwargs):
                 figs[i].canvas.draw()
 
                 plt.savefig(out_dir + fig_names[i].format(it / 1000), bbox_inches='tight')
-
+                if PLT_CLOSE == 1:
+                    plt.close()
             # Run evaluation functions
             for func in eval_funcs:
                 func(it, img_generator)
@@ -178,4 +179,4 @@ def began_disc(x, name, **kwargs):
         d_dec = DCGAN_G(n_in=dim_h, last_act=tf.sigmoid, bn=False)
 
         train_began(data, g_net, d_enc, d_dec, name=out_name, dim_z=dim_z, batch_size=args.batchsize, lr=args.lr,
-                    eval_funcs=[lambda it, gen: eval_images_naive(it, gen, data)])
+                    eval_funcs=[lambda it, gen: eval_images_naive(it, gen, data)])

common.py

@@ -9,13 +9,14 @@
 import argparse
 import numpy as np
 import io
-
+import matplotlib.cm as cm
 
 # Global configs
 PRNT_INTERVAL = 100
 EVAL_INTERVAL = 2000
+SHOW_FIG_INTERVAL = 1000
 SAVE_INTERVAL = 4000
-
+PLT_CLOSE = 1 #TODO set it to 1 for 1D exp as seaborn needs it
 DATASETS = ['mnist', 'celeba']
 
 
@@ -45,7 +46,7 @@ def create_dirs(name, g_name, d_name, hyperparams=None):
 def check_dataset_type(shape):
     assert(shape)
 
-    if len(shape) == 1:
+    if len(shape) == 1 or shape == (1,1):
         return 'synthetic'
     elif shape[2] == 1:
         assert(shape[0] == 28 and shape[1] == 28)
@@ -93,7 +94,13 @@ def scatter(samples, figId=None, retBytes=False, xlim=None, ylim=None):
         fig = plt.figure(figId)
         fig.clear()
 
+    n_gen = 8 #TODO
+    colors = cm.rainbow(np.linspace(0, 1, n_gen)) #TODO
+    colors = np.repeat(colors, len(samples[:,0])/n_gen, 0) #TODO
+
+    #plt.scatter(samples[:,0], samples[:,1], c = colors, alpha=0.1) #TODO
     plt.scatter(samples[:,0], samples[:,1], alpha=0.1)
+
     if xlim:
         plt.xlim(xlim[0], xlim[1])
     if ylim:
@@ -112,7 +119,7 @@ def scatter(samples, figId=None, retBytes=False, xlim=None, ylim=None):
 def parse_args(batchsize=128, lr=1e-5, additional_args=[]):
     parser = argparse.ArgumentParser()
 
-    parser.add_argument('--gpu', type=int, default=0)
+    parser.add_argument('--gpu', type=int, default=0)   #TODO
     parser.add_argument('--batchsize', type=int, default=batchsize)
     parser.add_argument('--datasets', choices=DATASETS, default=DATASETS[0])
     parser.add_argument('--lr', type=float, default=lr)
@@ -129,4 +136,4 @@ def parse_args(batchsize=128, lr=1e-5, additional_args=[]):
 def set_gpu(gpu_id):
     print "Override GPU setting: gpu={}".format(gpu_id)
     import os
-    os.environ["CUDA_VISIBLE_DEVICES"] = "{}".format(gpu_id)
+    os.environ["CUDA_VISIBLE_DEVICES"] = "{}".format(gpu_id)

datasets/data_celeba.py

@@ -121,4 +121,4 @@ def plot(self, img_generator, fig_id=None):
 
     for i in range(16):
         cv2.imshow('image', ims[i][:, :, (2,1,0)])
-        cv2.waitKey(0)
+        cv2.waitKey(0)

datasets/data_generator.lua

@@ -0,0 +1,71 @@
+#Uses  space separated specs.txt to output input.txt.
+#specs need to have a header row with four columns
+
+require 'torch'
+require 'nn'
+require 'optim'
+require 'pl'
+
+opt={
+    folder='',
+    data_name='',
+    num_samples=768000,
+    filename='specs.txt',
+    out_name='input.txt',
+    t7_filename='data.t7'
+}
+
+function tablelength(T)
+    local count = 0
+    for _ in pairs(T) do count = count + 1 end
+    return count
+end
+
+for k,v in pairs(opt) do opt[k] = tonumber(os.getenv(k)) or os.getenv(k) or opt[k] end
+print(opt)
+
+local d=data.read(paths.concat(opt.folder,opt.data_name,opt.filename))
+local ndim=1
+local colnames=d:column_names()
+if tablelength(colnames)==4 then
+    ndim=2
+elseif tablelength(colnames)==3 then
+    ndim=1
+end
+print('ndim '..ndim)
+
+local ncentres=tablelength( d:column_by_name(colnames[1]) )
+
+local centres=torch.Tensor(ncentres,ndim)
+local std_dev=torch.Tensor(ncentres)
+local densities=torch.Tensor(ncentres)
+
+for i=1,ncentres do
+    for j=1,ndim do
+        centres[i][j]=tonumber(d:column_by_name(colnames[j])[i])
+    end
+    std_dev[i]=tonumber(  d:column_by_name(colnames[ndim+1])[i] )
+    densities[i]=tonumber(  d:column_by_name(colnames[ndim+2])[i] )
+end
+
+paths.mkdir(paths.concat(opt.folder,opt.data_name))
+rand_indices=torch.multinomial(densities, opt.num_samples,true )
+local data=torch.Tensor(opt.num_samples,ndim)
+file=io.open(paths.concat(opt.folder,opt.data_name,opt.out_name) ,'w')
+io.output(file)
+for i=1,opt.num_samples do
+    local k=rand_indices[i]
+    local point=torch.Tensor(ndim)
+    for j=1,ndim do
+        point[j]=torch.normal(0,std_dev[k])+centres[k][j]
+    end
+    data[i]=point
+    if ndim==1 then
+        io.write(string.format('%d %f\n',0,point[1]))
+    elseif ndim==2 then
+        io.write(string.format('%d %f %f\n',0,point[1],point[2]))
+    end
+end
+io.close(file)
+
+torch.save( paths.concat(opt.folder,opt.data_name,opt.t7_filename),data)