Add files via upload

Uploading the latest version of ECGMOD

Author: pythonpanda2

ECGMOD/DKLTuner.py

@@ -65,17 +65,14 @@ def DKLmodel(x_train, x_test, y_train, y_test, numy=1, n_iter=500,\
     """
 
     if GP == "DKL" :
-        likelihood = gpytorch.likelihoods.GaussianLikelihood( noise_prior=gpytorch.priors.SmoothedBoxPrior(0.15, 1.5, sigma=0.5) ) 
+        likelihood = gpytorch.likelihoods.GaussianLikelihood( ) # noise_prior=gpytorch.priors.SmoothedBoxPrior(0.15, 1.5, sigma=0.5) 
         model = DKLRegressionModel(x_train, y_train, likelihood, kernel, x_train.size(-1), l1out, l2out, l3out, final)
 
     if torch.cuda.is_available():
         model = model.cuda()
         likelihood = likelihood.cuda()
 
 
-    model.train()
-    likelihood.train()
-
     #Define the parameters to optimize
     if GP == "DKL":
         optimizer = torch.optim.Adam([
@@ -85,9 +82,14 @@ def DKLmodel(x_train, x_test, y_train, y_test, numy=1, n_iter=500,\
     {'params': model.likelihood.parameters()},
 ], lr=lr)
 
+    scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=100, gamma=0.1)    
+
     # "Loss" for GPs - the marginal log likelihood
     mll = gpytorch.mlls.ExactMarginalLogLikelihood(likelihood, model)
     #training_iter = 500
+    
+    model.train()
+    likelihood.train()
 
     for i in range(n_iter):
         optimizer.zero_grad()
@@ -98,6 +100,7 @@ def DKLmodel(x_train, x_test, y_train, y_test, numy=1, n_iter=500,\
             if i % 100  == 0 or i == n_iter - 1:
                 print('Iter %d/%d - Loss: %.3f' % (i + 1, n_iter, loss.item()))
         optimizer.step()
+        scheduler.step()
 
     mse, r2, predictions = InferGPModel(likelihood, model, x_test, y_test)
     gpnoise = estimate_gpnoise(likelihood, verbose=True)

ECGMOD/GPmodels.py

@@ -162,15 +162,18 @@ def InferGPModel(likelihood, model, Xinfer, YTrue, scaleY=None):
     rmse, r2, predictions: float, float, tensor
     """ 
     from sklearn.metrics import mean_squared_error, r2_score
-    import numpy
-    if type(Xinfer) == numpy.ndarray:
+    import numpy as np 
+
+    if type(Xinfer) == np.ndarray:
         use_cuda = torch.cuda.is_available()
         dtype = torch.cuda.FloatTensor if use_cuda else torch.FloatTensor
         Xinfer = torch.from_numpy(Xinfer).type(dtype)    
         YTrue = torch.from_numpy(YTrue).type(dtype)
+
     # Set into eval mode
     model.eval()
     likelihood.eval()
+
     with torch.no_grad(), gpytorch.settings.fast_pred_var():
         predictions = likelihood(model(Xinfer))
         mean = predictions.mean
@@ -189,7 +192,7 @@ def InferGPModel(likelihood, model, Xinfer, YTrue, scaleY=None):
     mse = mean_squared_error(YTrue,mean)
     r2 = r2_score(YTrue,mean)
 
-    return numpy.sqrt(mse), r2, predictions
+    return np.sqrt(mse), r2, predictions
 
 
 def estimate_gpnoise(likelihood, verbose=True):

ECGMOD/multiT_preprocess.py

@@ -0,0 +1,56 @@
+def multiT_data_loader(jsonfile, NormalizeY=False, NormalizeX=False):
+    """
+    Input Args:
+    jsonfile  : JSON configuration filename (str) 
+    NormalizeY : default : False (Bool)
+    NormalizeX : default : False (Bool)
+    dataset    : default : 'SciAdv' (Str)
+    return:
+    train_x : train descriptor array (array)
+    train_y : train target array (array)
+    test_x  : test descriptor array (array)
+    test_y  : test target array (array)
+    numy : multi-task number(int)
+    scalerY : None if not normalized
+    scalerX : None if not normalized
+    """
+    import json, os
+    import pandas as pd
+    import numpy as np  
+    from sklearn.preprocessing import StandardScaler
+    
+    config_json = json.load(open(jsonfile))
+    #config_json = jsonfile
+    path = config_json['path']
+    train_file = config_json["train_file"]
+    test_file = config_json["test_file"]
+    x_slice = config_json["x_slice"] 
+    y_ind   = config_json["y_slice"] ## Assuming single task learning. Hence just an index.
+    
+    df_train =  np.load( os.path.join(path,train_file ) ).astype('float32')    
+    df_test =   np.load(os.path.join(path,test_file ) ).astype('float32')
+    
+    train_x, train_y  = df_train[:, x_slice[0] : x_slice[1] ], df_train[:, y_ind]
+    test_x, test_y    = df_test[:, x_slice[0] : x_slice[1] ], df_test[:, y_ind]
+    
+    if len(test_y.shape) == 1 :
+        train_y = train_y.reshape(-1, 1)
+        test_y = test_y.reshape(-1, 1)
+        numy = 1
+    else:
+        numy = test_y.shape[1]
+        
+    scalerY = None
+    scalerX = None
+    
+    if  NormalizeY:
+        scalerY = StandardScaler()
+        train_y  = scalerY.fit_transform( train_y )
+        test_y = scalerY.transform(test_y)
+
+    if NormalizeX:
+        scalerX = StandardScaler()
+        train_x  = scalerX.fit_transform( train_x )
+        test_x = scalerX.transform(test_x)
+    
+    return train_x, train_y, test_x, test_y, numy, scalerY, scalerX

ECGMOD/postprocess.py

@@ -1,4 +1,4 @@
-def plot_metric_write(train_size,err_list,r2_list, err_unscaled ,r2_unscaled , indlist):
+def plot_metric_write(train_size,err_list,r2_list, err_unscaled ,r2_unscaled , indlist, tag=None):
     '''
     Plot the metric/ write evolution history  over the trial using this function
     :param train_size: (list)
@@ -7,6 +7,7 @@ def plot_metric_write(train_size,err_list,r2_list, err_unscaled ,r2_unscaled , i
     :param err_unscaled : (list) Error rescaled to energy scale
     :param r2_unscaled : (list)
     :param indlist: (list) 
+    :param tag: (str) String to tag to output file name 
     Return type: figure object
     '''
     import matplotlib.pyplot as plt
@@ -52,9 +53,13 @@ def plot_metric_write(train_size,err_list,r2_list, err_unscaled ,r2_unscaled , i
     data = np.array([xarray, yarray, zarray, array_err_unscaled, array_r2_unscaled])
     data = data.T
 
-    datafile_path = "AL_history.dat"
     indexarray_path = "AL_train_Array.dat"
 
+    if tag == None:
+        datafile_path = "AL_history.dat"
+    elif tag != None:
+        datafile_path = "AL_history_{}.dat".format(tag)
+
     with open(datafile_path, 'w+') as datafile_id:
         np.savetxt(datafile_id, data, fmt=['%d','%f','%f','%f','%f'])