def test_dataframe_del(self):
rows = "row/"
cols = "col/"
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col/path1/"
col2 = "col/path2/"
matrix1 = np.arange(6).reshape(2,3)
matrix2 = np.arange(6).reshape(2,3)+10
matrix3 = np.arange(6).reshape(2,3)+20
matrix4 = np.arange(6).reshape(2,3)+30
matrix_row1 = np.hstack([matrix1,matrix2])
matrix_row2 = np.hstack([matrix3,matrix4])
matrix_all = np.vstack([matrix_row1,matrix_row2])
df = DataFrame()
M_df = DataFrame.from_matrix(matrix1)
df[row1,col1].set_matrix(matrix1)
df[row1,col2].set_matrix(matrix2)
df[row2,col1].set_matrix(matrix3)
df[row2,col2].set_matrix(matrix4)
assert (df[rows,cols].get_matrix()==matrix_all).all()
del df[row1,col1]
try:
df[row1,col1].get_matrix()
raise
except KeyError:
pass
print df._row_index
assert (df[rows,cols].get_matrix()==matrix4).all()
def test_json(self):
row = "row/"
col = "col/"
df = DataFrame()
matrix1 = np.arange(8).reshape(4,2)
df[row,col].set_matrix(matrix1)
d = json.loads(df.structure_to_json())
assert d["rows"][0]["directory"]
assert d["rows"][0]["query"] == "row/"
for i in range(4):
assert (d["rows"][0]["files"][i] ==
{"directory":False, 'query':str(i)})
assert d["cols"][0]["directory"]
assert d["cols"][0]["query"] == "col/"
for i in range(2):
assert (d["cols"][0]["files"][i] ==
{"directory":False, 'query':str(i)})
print d["row_index"]
print d["col_index"]
assert (d["row_index"] == {
"row/0":0,
"row/1":0,
"row/2":0,
"row/3":0
})
assert(d["col_index"] == {
"col/0":0,
"col/1":0,
})
assert(d["partitions"] == {
"(0, 0)" : 0
})
def test_dataframe_evictions(self):
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col/path1/"
all_rows = "row/"
matrix1 = np.arange(8).reshape(2,4)
matrix2 = np.arange(8).reshape(2,4)+2
df = DataFrame()
df[row1,col1] = DataFrame.from_matrix(matrix1)
df[row2,col1] = DataFrame.from_matrix(matrix2)
# load into cache
df[row1,col1].get_matrix()
assert(df[row1,col1]._is_cached())
df[row2,col1].get_matrix()
assert(df[row1,col1]._is_cached())
assert(df[row2,col1]._is_cached())
# assert that the combined value is correct
assert (df[all_rows,col1].get_matrix() ==
np.vstack([matrix1,matrix2])).all()
#assert cache membership
assert(df[all_rows,col1]._is_cached())
assert(not df[row1,col1]._is_cached())
assert(not df[row2,col1]._is_cached())
# attempt to modified a cached element
M = df[all_rows,col1].get_matrix()
M[0,:] = 0
df[all_rows,col1] = DataFrame.from_matrix(M)
# Assert the matrix is still in the cache
assert df[all_rows,col1]._is_cached()
# Assert that the new value is saved to cache
assert (df[all_rows,col1].get_matrix()==M).all()
# Pull an entry and evict all_rows
M1 = df[row1,col1].get_matrix()
M2 = df[row2,col1].get_matrix()
assert df[row1,col1]._is_cached()
assert df[row2,col1]._is_cached()
assert not df[all_rows,col1]._is_cached()
# assert that the pulled values match what they should
assert (np.vstack([M1,M2])==M).all()
assert (df[all_rows,col1].get_matrix()==M).all()
df[all_rows,col1][0:1,:].get_matrix()
assert (df[all_rows,col1][0:1,:]._is_cached())
# evict an entry that will change
row3 = "row/path3/"
matrix3 = np.arange(8).reshape(2,4)+4
df[row3,col1].set_matrix(matrix3)
assert (df[row3,col1].get_matrix()==matrix3).all()
def test_matrix_set(self):
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col/path1/"
col2 = "col/path2/"
matrix1 = np.arange(6).reshape(2,3)
df = DataFrame()
M_df = DataFrame.from_matrix(matrix1)
df[row1,col1].set_matrix(matrix1)
assert (df[row1,col1].get_matrix()==matrix1).all()
def test_simple_matrix_set(self):
df = DataFrame()
M = np.arange(4).reshape(2,2)
df["row/","col/"] = M
assert(df["row/","col/"].get_matrix() ==M).all()
df = DataFrame()
df["row/1","col/1"] = 0
df["row/1","col/2"] = 1
df["row/2","col/1"] = 2
df["row/2","col/2"] = 3
assert(df["row/","col/"].get_matrix() ==M).all()
df["row/","col/"] = 2*M
assert(df["row/","col/"].get_matrix() ==2*M).all()
def test_extend(self):
# Test extending an existing partition
row = "row/"
col = "col/"
df = DataFrame()
matrix1 = np.arange(8).reshape(4,2)
df[row,col].set_matrix(matrix1)
assert(df[row,col].shape==(4,2))
rows = [str(v) for v in range(6)]
cols = [str(v) for v in range(8)]
df[row,col]._extend(rows,cols)
assert(df[row,col].shape==(6,8))
M = df[row,col].get_matrix()
assert (M[0:4,0:2]==matrix1).all()
assert (M[5:,:]==0).all()
assert (M[:,2:]==0).all()
assert (df[row,col]._row_index.keys()==rows)
assert (df[row,col]._col_index.keys()==cols)
# Test extending an empty partition
row2 = "row2/"
col2 = "col2/"
df[row2,col2]._extend(rows,cols)
assert (df[row2,col2].shape == (6,8))
print df[row2,col2].get_matrix()
assert (df[row2,col2].get_matrix()==np.zeros((6,8))).all()
def test_matrix_property(self):
row = "row/"
col = "col/"
df = DataFrame()
matrix = np.arange(8).reshape(4,2)
df[row,col] = matrix
assert (df.r_matrix == matrix).all()
assert (df[row,col].r_matrix == matrix).all()
assert (df[row,col].rw_matrix == matrix).all()
df[row,col].rw_matrix = np.ones((4,2))
assert (df[row,col].rw_matrix == np.ones((4,2))).all()
try:
df[row,col].rw_matrix = 2
except ValueError:
pass
df[row,col].rw_matrix[:] = 2
assert (df[row,col].rw_matrix == 2*np.ones((4,2))).all()
df[row,col].rw_matrix = 2*matrix
assert (df[row,col].rw_matrix == 2*matrix).all()
def test_from_matrix(self):
matrix1 = np.ones((2,3))
df1 = DataFrame.from_matrix(matrix1, row_labels = ["a","b"])
df2 = DataFrame.from_matrix(matrix1, col_labels = ["a","b","c"])
df3 = DataFrame.from_matrix(matrix1,
col_labels = ["a","b", "c"],
row_labels = ["a","b"])
assert(df1.shape==matrix1.shape)
assert(df2.shape==matrix1.shape)
assert(df3.shape==matrix1.shape)
assert(df1.get_matrix().shape==matrix1.shape)
assert(df2.get_matrix().shape==matrix1.shape)
assert(df3.get_matrix().shape==matrix1.shape)
assert(df1.get_matrix()==matrix1).all()
assert(df2.get_matrix()==matrix1).all()
assert(df3.get_matrix()==matrix1).all()
def test_cache_rows_then_evict_all(self):
row = "row/"
col = "col/"
nrows = 4
matrix1 = np.arange(8).reshape(nrows,2)
df = DataFrame()
df[row,col] = DataFrame.from_matrix(matrix1)
for i in range(nrows):
assert (df[i,:].get_matrix()==matrix1[i,:]).all()
assert (df[:,:].get_matrix() == matrix1).all()
for i in range(0,nrows,2):
assert (df[i:i+2,:].get_matrix()==matrix1[i:i+2,:]).all()
assert (df[:,:].get_matrix() == matrix1).all()
def test_nonexistant_query(self):
row1 = "row/path1/"
row2 = "row/path2/"
row3 = "row/path3/"
col1 = "col/path1/"
col2 = "col/path2/"
col3 = "col/path3/"
matrix1 = np.arange(6).reshape(2,3)
df = DataFrame()
df[row1,col1] = DataFrame.from_matrix(matrix1)
df[row2,col2]
df[row3,col3] = dreaml.transformations.Identity(df[row1,col1])
assert df[row2,col2].empty()
assert df[row2,col3].empty()
assert df[row3,col2].empty()
assert not df[row1,col1].empty()
assert not df[row3,col3].empty()
def test_pwd(self):
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col/path1/"
col2 = "col/path2/"
matrix1 = np.arange(6).reshape(2,3)
matrix2 = np.arange(6).reshape(2,3)+4
df = DataFrame()
df[row1,col1].set_matrix(matrix1)
df[row2,col2].set_matrix(matrix2)
assert(df.pwd() == ("",""))
assert(df[row1,col2].pwd() == (row1,col2))
assert(df["row/","col/"].pwd() == ("row/","col/"))
assert(df["row/","col/"]["path1/",:].pwd() == ("row/path1/","col/"))
assert(df["row/","col/"][0,1:2]["path1/",:].pwd() ==
("row/path1/","col/"))
assert(df["row/path1/0","col/path2/0"].pwd() ==
("",""))
assert(df["row/","col/"]["path2/0","path1/0"].pwd() ==
("row/","col/"))
def test_dataframe_update_propogation(self):
row1 = "row/path1/"
row11 = "row/path1/sub1/"
row12 = "row/path1/sub2/"
row2 = "row/path2/"
cols = "col/"
df = DataFrame()
matrix1 = np.arange(8).reshape(2,4)
df[row11,cols] = DataFrame.from_matrix(matrix1)
df[row2,cols] = dreaml.transformations.Identity(df[row1,cols])
# Check a few equalities from directory indexing
assert (df[row11,cols].get_matrix()==df[row1,cols].get_matrix()).all()
assert (df[row1,cols].get_matrix()==matrix1).all()
assert (df[row2,cols].get_matrix()==matrix1).all()
# validate some simple graph properties
h1 = df[row1,cols].hash()
h2 = df[row2,cols].hash()
assert df._graph.node[h1]["status"] == df.STATUS_GREEN
assert df._graph.node[h2]["status"] == df.STATUS_GREEN
matrix2 = np.arange(8).reshape(2,4) + 5
df[row12,cols] = DataFrame.from_matrix(matrix2)
# Check that cache entries have been properly invalidated
assert not df[row1,cols]._is_cached()
# Check that set matrices are correct
assert (df[row11,cols].get_matrix() == matrix1).all()
assert df[row11,cols]._is_cached()
assert (df[row12,cols].get_matrix() == matrix2).all()
assert df[row12,cols]._is_cached()
# check that total matrix is correct
matrix3 = np.vstack([matrix1,matrix2])
assert (df[row1,cols].get_matrix()==matrix3).all()
# assert that the two matrices are equal
assert (df[row1,cols].shape==df[row2,cols].shape)
assert (df[row1,cols].get_matrix()==df[row2,cols].get_matrix()).all()
def test_sequentialy_matrix_set(self):
df = DataFrame()
M1 = np.arange(4).reshape(2,2)
M2 = np.arange(4).reshape(2,2)*2
M3 = np.arange(4).reshape(2,2)*3
df["row1/","col1/"] = M1
df["row1/","col2/"] = M2
df["row2/","col1/"] = M3
assert (df["row1/","col1/"].r_matrix == M1).all()
assert (df["row1/","col2/"].r_matrix == M2).all()
assert (df["row2/","col1/"].r_matrix == M3).all()
def test_single_set(self):
df = DataFrame()
M = np.arange(4).reshape(2,2)
df["row/1","col/1"] = 0
df["row/1","col/2"] = 1
df["row/2","col/1"] = 2
assert not df["row/2","col/2"].empty()
df["row/2","col/2"].set_matrix(3)
assert(df["row/","col/"].shape==(2,2))
assert(df["row/","col/"].get_matrix() == M).all()
df["a/","a/"].set_matrix(3)
assert(df["a/","a/"].shape==(1,1))
assert(df["a/","a/"].get_matrix()==3).all()
def test_dataframe_hierarchy(self):
# Add a new row entry within row directory
rows = "rows/"
row1 = "rows/path1/"
row2 = "rows/path2/"
cols = "cols/"
col1 = "cols/p1/"
col2 = "cols/p2/"
matrix1 = np.arange(6).reshape(2,3)
matrix2 = np.arange(6).reshape(2,3)+8
df = DataFrame()
df[row1,col1] = DataFrame.from_matrix(matrix1)
assert (df[row1,col1].get_matrix() == matrix1).all()
assert (df[rows,col1].get_matrix() == matrix1).all()
assert df[rows,col1]._is_cached()
df_both_rows = df[rows,col1]
df[row2,col1] = DataFrame.from_matrix(matrix2)
# Previous query should be evicted at this point
assert df[row1,col1]._is_df_cached()
assert df[row2,col1]._is_df_cached()
assert not df_both_rows._is_df_cached()
print df._cache
assert not df_both_rows._is_cached()
assert (df[row2,col1].get_matrix() == matrix2).all()
# Check fetching everything
matrix3 = np.vstack([matrix1,matrix2])
print df[rows,col1].get_matrix()
print matrix3
print df[rows,col1]._row_index
assert (df[rows,col1].shape == matrix3.shape)
assert (df[rows,col1].get_matrix() == matrix3).all()
def test_df_cache(self):
rows = "row/"
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col1/"
col2 = "col2/"
matrix1 = np.arange(6).reshape(2,3)
matrix2 = np.arange(6).reshape(2,3)+2
df = DataFrame()
df[row1,col1].set_matrix(matrix1)
# Put rows,col1 into the dataframe cache
assert (df[rows,col1].get_matrix() == matrix1).all()
df[row2,col2].set_matrix(matrix2)
assert (df[rows,col1].shape==(4,3))
assert (df[rows,col1].get_matrix()==
np.vstack([matrix1,np.zeros((2,3))])).all()
def test_dataframe_empty(self):
df = DataFrame()
df["row/1","col/1"] = 0
assert not df["row/1","col/1"].empty()
try:
df["row/1","col/bad"]
except KeyError:
pass
try:
df["row/bad","col/1"]
except KeyError:
pass
try:
df["row/bad","col/bad"]
except KeyError:
pass
assert not df["row/","col/"].empty()
assert df["bad/","col/"].empty()
assert df["row/","bad/"].empty()
assert df["bad/","bad/"].empty()
def test_multi_dataframe_set(self):
df1 = DataFrame()
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col/path1/"
col2 = "col/path2/"
matrix1 = np.arange(6).reshape(2,3)
matrix2 = np.arange(6).reshape(2,3)+2
matrix3 = np.arange(6).reshape(2,3)+4
matrix4 = np.arange(6).reshape(2,3)+6
df1[row1,col1].set_matrix(matrix1)
df1[row1,col2].set_matrix(matrix2)
df1[row2,col1].set_matrix(matrix3)
df1[row2,col2].set_matrix(matrix4)
df2 = DataFrame()
df2["parentrow/","parentcol/"] = df1
assert (df2.get_matrix() == df1.get_matrix()).all()
def _start_empty():
df = DataFrame()
f = lambda a: create_app(a).run()
df.web_thread = Thread(target=f, args=(df,))
df.web_thread.start()
def _start_empty():
df = DataFrame()
f = lambda a: create_app(a).run()
df.web_thread = Thread(target=f, args=(df, ))
df.web_thread.start()
def test_dataframe_basic(self):
print "kdfd"
row1 = "row/path1/"
row2 = "row/path2/"
col1 = "col/path1/"
col2 = "col/path2/"
matrix1 = np.arange(6).reshape(2,3)
df = DataFrame()
# After inserting one block, other entries should not work
df[row1,col1] = DataFrame.from_matrix(matrix1)
assert (df[row1,col1].get_matrix() == matrix1).all()
try:
# missing column
df[row1,col2].get_matrix()
raise
except KeyError:
pass
try:
# missing row
df[row2,col1].get_matrix()
raise
except KeyError:
pass
try:
# missing both
df[row1,col2].get_matrix()
raise
except KeyError:
pass
# Try overwriting the written matrix:
matrix2 = np.arange(6).reshape(2,3)+4
df[row1,col1] = DataFrame.from_matrix(matrix2)
assert (df[row1,col1].get_matrix() == matrix2).all()
# try indexing with slices
assert(df[row1,col1][0,:].get_matrix()==matrix2[0,:]).all()
# try subtracting 1 from a row of dataframe partition
matrix3 = matrix2
matrix3[0,:] -= 1
tmp = df[row1,col1][0,:].get_matrix()
tmp -= 1
df[row1,col1][0,:] = DataFrame.from_matrix(tmp)
assert (df[row1,col1].get_matrix()==matrix3).all()
# Test setting a single element
df[1,1] = 5
assert(df[1,1].get_matrix()[0,0] == 5)
assert(df[1,1]._is_cached())
# Test use of set_matrix on a single query
matrix3 = np.arange(6).reshape(2,3)
df[row1,col1].set_matrix(matrix3+5)
assert not df[1,1]._is_cached()
assert(df[row1,col1].get_matrix()==(matrix3+5)).all()
assert df[row1,col1]._is_cached()
# Test set_matrix on a nested query
matrix4 = np.array([[42]])
df[row1,col1][0,1].set_matrix(matrix4)
assert (df[row1,col1][0,1].get_matrix()==matrix4).all()
