adding plots pdf, update existing code and news files

norm_data.py

+import pandas as pd
+import numpy as np
+import os
+
+def normalize_time_series (radar_file, opt_file, norm="meanstd", radar_bands=["VV","VH"], 
+                            opt_bands=["B2","B3","B4","B8","B5","B6","B7","B8A","B11","B12"], 
+                            opt_indices = ["NDVI","NDWI","EVI","MSAVI2","GDVI","CIGreen","CIRedEdge"]):
+    
+    radar_df = pd.read_csv(radar_file)
+    radar_array = None
+    for band in radar_bands :
+        columns = [col for col in radar_df.columns if col.startswith(band) and col.endswith("Mean")]
+        columns.sort()
+        ts_values = radar_df[columns].values
+        if norm == "meanstd":
+            ts_values = (ts_values - ts_values.mean()) / ts_values.std()
+        elif norm == "minmax":
+            ts_values = (ts_values - ts_values.min()) / (ts_values.max() - ts_values.min())
+        if radar_array is None :
+            radar_array = ts_values
+        else :
+            radar_array = np.hstack((radar_array,ts_values))
+    
+    n_timestamps = len(columns)
+    n_bands = len(radar_bands)
+    radar_seq = None
+
+    for i in range(n_timestamps):
+        lst = []
+        for j in range(n_bands):
+            lst.append(radar_array[:,i+j*n_timestamps]) 
+
+        if radar_seq is None :
+            radar_seq = np.stack(lst, axis=1)
+        else :
+            radar_seq = np.hstack((radar_seq,np.stack(lst, axis=1)))
+
+    opt_df = pd.read_csv(opt_file)
+    opt_array = None
+    for band in opt_bands :
+        columns = [col for col in opt_df.columns if col.split("_")[0]==band and col.endswith("Mean")]
+        columns.sort()
+        ts_values = opt_df[columns].values
+        if norm == "meanstd":
+            ts_values = (ts_values - ts_values.mean()) / ts_values.std()
+        elif norm == "minmax":
+            ts_values = (ts_values - ts_values.min()) / (ts_values.max() - ts_values.min())
+        if opt_array is None :
+            opt_array = ts_values
+        else :
+            opt_array = np.hstack((opt_array,ts_values))
+
+    n_timestamps = len(columns)
+    n_bands = len(opt_bands)
+    opt_seq = None
+
+    for i in range(n_timestamps):
+        lst = []
+        for j in range(n_bands):
+            lst.append(opt_array[:,i+j*n_timestamps]) 
+
+        if opt_seq is None :
+            opt_seq = np.stack(lst, axis=1)
+        else :
+            opt_seq = np.hstack((opt_seq,np.stack(lst, axis=1)))
+
+    indices_df = pd.read_csv(opt_file)
+    indices_array = None
+    for band in opt_indices :
+        columns = [col for col in indices_df.columns if col.startswith(band) and col.endswith("Mean")]
+        columns.sort()
+        ts_values = indices_df[columns].values
+        if norm == "meanstd":
+            ts_values = (ts_values - ts_values.mean()) / ts_values.std()
+        elif norm == "minmax":
+            ts_values = (ts_values - ts_values.min()) / (ts_values.max() - ts_values.min())
+        if indices_array is None :
+            indices_array = ts_values
+        else :
+            indices_array = np.hstack((indices_array,ts_values))
+
+    n_timestamps = len(columns)
+    n_bands = len(opt_indices)
+    indices_seq = None
+
+    for i in range(n_timestamps):
+        lst = []
+        for j in range(n_bands):
+            lst.append(indices_array[:,i+j*n_timestamps]) 
+
+        if indices_seq is None :
+            indices_seq = np.stack(lst, axis=1)
+        else :
+            indices_seq = np.hstack((indices_seq,np.stack(lst, axis=1)))
+    
+    ptrn = os.path.basename(radar_file).split('_')[0]+"_"+os.path.basename(radar_file).split('_')[1]
+    np.save("./data/{}_rad_seq.npy".format(ptrn),radar_seq)
+    np.save("./data/{}_opt_seq.npy".format(ptrn),opt_seq)
+    np.save("./data/{}_indices_seq.npy".format(ptrn),indices_seq)
+
+    try:
+        rdt_df = radar_df.merge(opt_df[["ID"]],on="ID")
+        rdt = rdt_df[["Rdt_s"]].values
+        np.save("./data/{}_yields.npy".format(ptrn),rdt)
+    except Exception as error : 
+        print (error)
+    
+
+
+if __name__ == '__main__' :
+
+    # Niakhar 2017
+    radar_file = "./data/niakhar_2017_radar_notree.csv"
+    opt_file = "./data/niakhar_2017_opt_gapf_notree.csv"
+    normalize_time_series(radar_file,opt_file,norm="meanstd")
+
+    # Niakhar 2018
+    radar_file = "./data/niakhar_2018_radar_notree.csv"
+    opt_file = "./data/niakhar_2018_opt_gapf_notree.csv"
+    normalize_time_series(radar_file,opt_file,norm="meanstd")
+
+    # Nioro 2018
+    radar_file = "./data/nioro_2018_radar_notree.csv"
+    opt_file = "./data/nioro_2018_opt_gapf_notree.csv"
+    normalize_time_series(radar_file,opt_file,norm="meanstd")
\ No newline at end of file

split_cross_validation.py

+import random
+from datetime import datetime
+import numpy as np
+import os
+
+def writeSplit (idx, data, outFileName):
+    fold = data[idx]
+    np.save(outFileName, fold)
+    
+def split_data (radar_seq,opt_seq,indices_seq,yields,n_folds=3,n_random=10) :
+    '''
+    '''
+    ptrn = os.path.basename(radar_seq).split("_")[0]+"_"+os.path.basename(radar_seq).split("_")[1]
+    if not os.path.exists (os.path.join("splits",ptrn)):
+        os.makedirs(os.path.join("splits",ptrn))
+    print (ptrn)
+
+    y = np.load(yields)
+    radar = np.load(radar_seq)
+    opt = np.load(opt_seq)
+    indices = np.load(indices_seq)
+
+    idx = np.arange(y.shape[0])
+    n_samples = int(y.shape[0]/n_folds)
+    
+    dt = datetime.now()
+    random.seed(dt.microsecond)
+
+    for j in range(n_random) :
+        for i in range(n_folds):
+            if i==0:
+                random.shuffle(idx)
+                random.shuffle(idx)
+
+            test_samples = idx[i*n_samples:(i+1)*n_samples]
+            test_idx = np.where(np.isin(idx,test_samples))
+            train_idx = np.where(np.isin(idx,test_samples,invert=True))
+
+            writeSplit(train_idx, radar, os.path.join("splits",ptrn,"train_radar_{}_fold-{}.npy".format(j+1,i+1)))
+            writeSplit(test_idx, radar, os.path.join("splits",ptrn,"test_radar_{}_fold-{}.npy".format(j+1,i+1)))
+
+            writeSplit(train_idx, opt, os.path.join("splits",ptrn,"train_opt_{}_fold-{}.npy".format(j+1,i+1)))
+            writeSplit(test_idx, opt, os.path.join("splits",ptrn,"test_opt_{}_fold-{}.npy".format(j+1,i+1)))
+
+            writeSplit(train_idx, indices, os.path.join("splits",ptrn,"train_indices_{}_fold-{}.npy".format(j+1,i+1)))
+            writeSplit(test_idx, indices, os.path.join("splits",ptrn,"test_indices_{}_fold-{}.npy".format(j+1,i+1)))
+
+            writeSplit(train_idx, y, os.path.join("splits",ptrn,"train_yields_{}_fold-{}.npy".format(j+1,i+1)))
+            writeSplit(test_idx, y, os.path.join("splits",ptrn,"test_yields_{}_fold-{}.npy".format(j+1,i+1)))
+
+        print ("========= split {} =========".format(j+1))
+        print ("%d-folds : %d, %d over %d" % (n_folds,train_idx[0].shape[0], test_idx[0].shape[0], len(idx)))
+
+if __name__ == "__main__":
+
+    # Niakhar 2017
+    radar_seq = "./data/niakhar_2017_rad_seq.npy"
+    opt_seq = "./data/niakhar_2017_opt_seq.npy"
+    indices_seq = "./data/niakhar_2017_indices_seq.npy"
+    yields = "./data/niakhar_2017_yields.npy"
+
+    split_data (radar_seq,opt_seq,indices_seq,yields)
+
+    # Niakhar 2018
+    radar_seq = "./data/niakhar_2018_rad_seq.npy"
+    opt_seq = "./data/niakhar_2018_opt_seq.npy"
+    indices_seq = "./data/niakhar_2018_indices_seq.npy"
+    yields = "./data/niakhar_2018_yields.npy"
+
+    split_data (radar_seq,opt_seq,indices_seq,yields)
+
+    # Nioro 2018
+    radar_seq = "./data/nioro_2018_rad_seq.npy"
+    opt_seq = "./data/nioro_2018_opt_seq.npy"
+    indices_seq = "./data/nioro_2018_indices_seq.npy"
+    yields = "./data/nioro_2018_yields.npy"
+
+    split_data (radar_seq,opt_seq,indices_seq,yields)
\ No newline at end of file