From 404cbf59f70f1fd730326e7c3c1ce0a3921beb96 Mon Sep 17 00:00:00 2001 From: Violeta Gonzalez-Perez Date: Fri, 31 May 2019 16:24:18 +0100 Subject: [PATCH] Formatted code with black: novonix_add.py and test_novonix_add.py --- .flake8 | 5 - .pre-commit-config.yaml | 1 + preparenovonix/.flake8 | 5 + preparenovonix/novonix_add.py | 988 +++++++++++++++++------------ preparenovonix/test_novonix_add.py | 25 +- pyproject.toml | 16 - 6 files changed, 598 insertions(+), 442 deletions(-) delete mode 100644 .flake8 create mode 100644 preparenovonix/.flake8 delete mode 100644 pyproject.toml diff --git a/.flake8 b/.flake8 deleted file mode 100644 index d9ad0b4..0000000 --- a/.flake8 +++ /dev/null @@ -1,5 +0,0 @@ -[flake8] -ignore = E203, E266, E501, W503, F403, F401 -max-line-length = 79 -max-complexity = 18 -select = B,C,E,F,W,T4,B9 diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml index e97f183..aad7a51 100644 --- a/.pre-commit-config.yaml +++ b/.pre-commit-config.yaml @@ -4,6 +4,7 @@ repos: hooks: - id: black language_version: python3.7 + files: ^preparenovonix/ - repo: https://github.com/pre-commit/pre-commit-hooks rev: v1.2.3 hooks: diff --git a/preparenovonix/.flake8 b/preparenovonix/.flake8 new file mode 100644 index 0000000..c661bb4 --- /dev/null +++ b/preparenovonix/.flake8 @@ -0,0 +1,5 @@ +[flake8] +ignore = E203, E266, E401, E501, W503, F403, F401, C901 +max-line-length = 88 +max-complexity = 18 +select = B,C,E,F,W,T4,B9 diff --git a/preparenovonix/novonix_add.py b/preparenovonix/novonix_add.py index 9390c74..7f1ef3d 100644 --- a/preparenovonix/novonix_add.py +++ b/preparenovonix/novonix_add.py @@ -1,58 +1,72 @@ import sys, os import numpy as np from pathlib import Path -from shutil import move,copy +from shutil import move, copy """A python module to add information to the Novonix data. .. moduleauthor:: Violeta Gonzalez-Perez """ -eps = 1.e-5 +eps = 1.0e-5 # Novonix columns -col_step = 'Step Number' -col_tstep = 'Step Time (h)' -col_capacity = 'Capacity (Ah)' +col_step = "Step Number" +col_tstep = "Step Time (h)" +col_capacity = "Capacity (Ah)" -state_col = 'State (0=Start 1=Regular 2=End -1=Single Measurement)' -loop_col = 'Loop number' -line_col = 'Protocol Line (it refers to the reduced protocol)' +state_col = "State (0=Start 1=Regular 2=End -1=Single Measurement)" +loop_col = "Loop number" +line_col = "Protocol Line (it refers to the reduced protocol)" # Commands in Protocols -increment1 = 'Increment the cycle counter' -increment2 = 'Increment cycle counter' - -ignore = ['End storage','End charge','End discharge','End repeat', - increment1,increment2,'End increment', - 'Trip conditions','End trip conditions', - 'Save conditions','End save conditions', - 'Operating limits','End operating limits', - 'Emergency limits','End emergency limits'] - -com_prot = ['Open_circuit_storage', - 'Constant_current_charge', - 'Constant_current_discharge', - 'CC-CV_charge', - 'CC-CV_discharge', - 'Repeat'] +increment1 = "Increment the cycle counter" +increment2 = "Increment cycle counter" + +ignore = [ + "End storage", + "End charge", + "End discharge", + "End repeat", + increment1, + increment2, + "End increment", + "Trip conditions", + "End trip conditions", + "Save conditions", + "End save conditions", + "Operating limits", + "End operating limits", + "Emergency limits", + "End emergency limits", +] + +com_prot = [ + "Open_circuit_storage", + "Constant_current_charge", + "Constant_current_discharge", + "CC-CV_charge", + "CC-CV_discharge", + "Repeat", +] # Novonix step values OCV = 0 CCc = 1 -CCd = 2 #in some cases this is also set to 1 +CCd = 2 # in some cases this is also set to 1 CCCV_CCc = 7 CCCV_CVc = 8 CCCV_CCd = 9 CCCV_CVd = 10 -com_val1 = [OCV,CCc,CCd,CCCV_CCc,CCCV_CCd] -com_val2 = [None,None,CCc,CCCV_CVc,CCCV_CVd] +com_val1 = [OCV, CCc, CCd, CCCV_CCc, CCCV_CCd] +com_val2 = [None, None, CCc, CCCV_CVc, CCCV_CVd] + +numberstr = ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "-"] -numberstr = ['0','1','2','3','4','5','6','7','8','9','-'] def select_com_val(index): - ''' + """ Establish the selection statement for the one or two possible command values (com_val#) @@ -72,16 +86,17 @@ def select_com_val(index): >>>> prep.select_com_val(2) 'np.logical_or(step == com_val1[index], step == com_val2[index])' - ''' - if (com_val2[index] is None): - sel = 'step == com_val1[index]' + """ + if com_val2[index] is None: + sel = "step == com_val1[index]" else: - sel = 'np.logical_or(step == com_val1[index], step == com_val2[index])' + sel = "np.logical_or(step == com_val1[index], step == com_val2[index])" return sel - + + def isnovonix(infile): """ - Given a data file, check if it exists and + Given a data file, check if it exists and if looks like a Novonix file, allowing for blank lines and commas after the commands due to having open the file in Excel @@ -104,67 +119,86 @@ def isnovonix(infile): """ answer = True - + # Test if the file exists - if(not os.path.isfile(infile)): + if not os.path.isfile(infile): answer = False - print('STOP function isnovonix \n'+\ - 'REASON Input file not found: '+str(infile)+' \n') + print( + "STOP function isnovonix \n" + + "REASON Input file not found: " + + str(infile) + + " \n" + ) return answer else: - with open(infile,'r') as ff: + with open(infile, "r") as ff: # Read until different header statement - keyws = ['Summary','Novonix','Protocol','Data'] + keyws = ["Summary", "Novonix", "Protocol", "Data"] for keyw in keyws: for line in ff: if line.strip(): char1 = line.strip()[0] - if (char1 in numberstr): + if char1 in numberstr: answer = False - print('STOP function isnovonix \n'+\ - 'REASON Reached the end of the input file \n'+\ - ' '+str(infile)+', \n'+\ - ' without the '+keyw+' entry.') + print( + "STOP function isnovonix \n" + + "REASON Reached the end of the input file \n" + + " " + + str(infile) + + ", \n" + + " without the " + + keyw + + " entry." + ) return answer else: - if (keyw in line): + if keyw in line: break # Read until the data starts for line in ff: if line.strip(): char1 = line.strip()[0] - if (char1 in numberstr): + if char1 in numberstr: break else: last_line = line.strip() - # From the data header, read the column names - colnames = last_line.split(',') + # From the data header, read the column names + colnames = last_line.split(",") # Remove triling blancks and end of lines - colnames = [x.strip() for x in colnames] + colnames = [x.strip() for x in colnames] # Check the existance of the "Step Number" column - if (col_step not in colnames): + if col_step not in colnames: answer = False - print('STOP function isnovonix \n'+\ - 'REASON No "Step Number" colum found in input file \n'+\ - ' '+str(infile)+' \n') + print( + "STOP function isnovonix \n" + + 'REASON No "Step Number" colum found in input file \n' + + " " + + str(infile) + + " \n" + ) return answer - + # Check the existance of the "Step time" column - if (col_tstep not in colnames): + if col_tstep not in colnames: answer = False - print('STOP function isnovonix \n'+\ - 'REASON No "Step Time" colum found in input file \n'+\ - ' '+str(infile)+' \n') + print( + "STOP function isnovonix \n" + + 'REASON No "Step Time" colum found in input file \n' + + " " + + str(infile) + + " \n" + ) return answer return answer -def icolumn(infile,column_name): + +def icolumn(infile, column_name): """ Given a Novonix data file, find the column position of the given column_name @@ -184,7 +218,7 @@ def icolumn(infile,column_name): Returns -------- icol : integer - Position of the column with name 'column_name', + Position of the column with name 'column_name', in the range form 0 to the number of columns - 1 Examples @@ -196,32 +230,34 @@ def icolumn(infile,column_name): """ icol = -1 - + # Check if the file has the expected structure for a Novonix file answer = isnovonix(infile) - if (not answer): sys.exit('STOP Input not from Novonix, {}'.format(file_to_open)) + if not answer: + sys.exit("STOP Input not from Novonix, {}".format(infile)) - with open(infile,'r') as ff: + with open(infile, "r") as ff: # Read until the line with [Data] for line in ff: - if ("[Data]" in line): + if "[Data]" in line: break # From the data header, read the column names - line = ff.readline() - colnames = line.split(',') - + line = ff.readline() + colnames = line.split(",") + # Find the position of the given column name - ii = 0 + ii = 0 for col in colnames: - if (column_name == col.strip()): + if column_name == col.strip(): icol = ii return icol ii += 1 - + return icol -def read_column(infile,column_name,outtype='float'): + +def read_column(infile, column_name, outtype="float"): """ Given a Novonix data file, read a column as an array of the type given in the variable astype. @@ -246,7 +282,8 @@ def read_column(infile,column_name,outtype='float'): --------- >>>> import pycode.novonix_add as prep - >>>> col = prep.read_column('example_data/example_data_prep.csv','Step Number',outtype='int') + >>>> col = prep.read_column('example_data/example_data_prep.csv', + 'Step Number',outtype='int') >>>> print(col[0]) 0 @@ -254,40 +291,42 @@ def read_column(infile,column_name,outtype='float'): # Check if the file has the expected structure for a Novonix file answer = isnovonix(infile) - if (not answer): sys.exit('STOP Input not from Novonix, {}'.format(file_to_open)) + if not answer: + sys.exit("STOP Input not from Novonix, {}".format(infile)) # Initialise empty list column_data = [] - with open(infile,'r') as ff: + with open(infile, "r") as ff: # Read until the line with [Data] for line in ff: - if ("[Data]" in line): + if "[Data]" in line: break # Read the column names - line = ff.readline() - colnames = line.split(',') + line = ff.readline() + colnames = line.split(",") # Find the position of the given column name ii = -1 for col in colnames: ii += 1 - if (column_name == col.strip()): + if column_name == col.strip(): icolumn = ii break # Read the column of interest for line in ff: - val = line.split(',')[icolumn].rstrip() + val = line.split(",")[icolumn].rstrip() column_data.append(val) # Transform the list into a numpy array column_data = np.array(column_data) - column = column_data.astype(getattr(np,outtype)) + column = column_data.astype(getattr(np, outtype)) return column + def cleannovonix(infile): """ Given a Novonix file remove blank lines, correct the header @@ -309,37 +348,40 @@ def cleannovonix(infile): >>>> prep.cleannovonix('example_data/example_data_prep.csv') """ - summary = '[Summary]' - + summary = "[Summary]" + # Check if there are unfinished tests ntests = 0 - with open(infile,'r') as ff: + with open(infile, "r") as ff: for line in ff: - if line.strip(): #Jump empty lines - if (summary in line): + if line.strip(): # Jump empty lines + if summary in line: # Count failed tests for each start of file ntests += 1 - if (ntests > 1): + if ntests > 1: # Find the capacity colum - icapacity = icolumn(infile,col_capacity) + icapacity = icolumn(infile, col_capacity) # Start reading the last test # Remove blank lines if present in the header - itest = 0 ; header = [] ; lastline = ' ' - last_capacity = 0. - with open(infile,'r') as ff: + itest = 0 + header = [] + lastline = " " + last_capacity = 0.0 + with open(infile, "r") as ff: for line in ff: if line.strip(): - if (summary in line): + if summary in line: itest += 1 - - if(ntests > 1 and itest > 1): - #Add last capacity of each failed test - last_capacity = last_capacity + float(lastline.split(',')[icapacity]) - - if (itest == ntests): - line1 = line.strip() - header.append(summary+' \n') + + if ntests > 1 and itest > 1: + # Add last capacity of each failed test + last_capacity = last_capacity + float( + lastline.split(",")[icapacity] + ) + + if itest == ntests: + header.append(summary + " \n") break lastline = line @@ -347,88 +389,93 @@ def cleannovonix(infile): for line in ff: if line.strip(): char1 = line.strip()[0] - if(char1 == '['): - cleanhead = line.split(']') - header.append(cleanhead[0]+'] \n') - if (cleanhead[0] == '[Data'): + if char1 == "[": + cleanhead = line.split("]") + header.append(cleanhead[0] + "] \n") + if cleanhead[0] == "[Data": break else: - header.append(line) + header.append(line) # From the data header, read the column names for line in ff: if line.strip(): - colnames = line.split(',') + colnames = line.split(",") break # Remove triling blancks and end of lines - colnames = [x.strip() for x in colnames] + colnames = [x.strip() for x in colnames] # Check that the number of data columns matches the header line_data1 = ff.readline() - data = line_data1.split(',') + data = line_data1.split(",") diff = len(data) - len(colnames) - if(diff>0): + if diff > 0: # Add dum headers - dums = '' + dums = "" for idiff in range(diff): - dums = dums+',dum'+str(idiff) - - new_head = str(line.rstrip())+dums+' \n' + dums = dums + ",dum" + str(idiff) + + new_head = str(line.rstrip()) + dums + " \n" header.append(new_head) - elif(diff<0): - sys.exit('STOP function cleannovonix \n'+\ - 'REASON less data columns than header names \n'+\ - ' '+str(infile)+' \n') + elif diff < 0: + sys.exit( + "STOP function cleannovonix \n" + + "REASON less data columns than header names \n" + + " " + + str(infile) + + " \n" + ) else: header.append(line) # Create a temporary file without blanck lines # and new header if needed - tmp_file = 'tmp.csv' - with open(tmp_file, 'w') as tf: + tmp_file = "tmp.csv" + with open(tmp_file, "w") as tf: for item in header: tf.write(str(item)) # Append the data - with open(tmp_file, 'a') as tf: + with open(tmp_file, "a") as tf: # Write the first data row - if (ntests > 1): + if ntests > 1: # Modify the Capacity column in case of failed tests new_capacity = float(data[icapacity]) + float(last_capacity) data[icapacity] = str(new_capacity) new_line = data[0] for col in data[1:]: - new_line = new_line+','+col + new_line = new_line + "," + col tf.write(new_line) else: tf.write(line_data1) # Write the rest of the data for line in ff: - if (ntests > 1): + if ntests > 1: # Modify the Capacity column in case of failed tests - columns = line.split(',') + columns = line.split(",") new_capacity = float(columns[icapacity]) + float(last_capacity) columns[icapacity] = str(new_capacity) new_line = columns[0] for col in columns[1:]: - new_line = new_line+','+col + new_line = new_line + "," + col tf.write(new_line) else: tf.write(line) # Replace the input file with the new one - move(tmp_file,infile) + move(tmp_file, infile) return -def novonix_add_state(infile,verbose=False): + +def novonix_add_state(infile, verbose=False): """ - Given a cleaned Novonix data file, it adds a 'State' column, + Given a cleaned Novonix data file, it adds a 'State' column, which mimimcs Basytec format with: 0 = Start of a measurement type (charge/discharge, etc) @@ -457,126 +504,152 @@ def novonix_add_state(infile,verbose=False): >>>> prep.novonix_add_state('example_data/example_data_prep.csv',verbose=True) The file example_data/example_data_prep.csv already has a State column """ - + # Check if the file already has the new State column - icol = icolumn(infile,state_col) - if (icol>-1): + icol = icolumn(infile, state_col) + if icol > -1: if verbose: - print('The file {} already has a State column'.format(infile)) + print("The file {} already has a State column".format(infile)) return # Find the Step Number column - icol = icolumn(infile,col_step) - icolt = icolumn(infile,col_tstep) + icol = icolumn(infile, col_step) + icolt = icolumn(infile, col_tstep) # The file need to have a 'State' column - header = [] ; fw = 'fw' - with open(infile,'r') as ff: + header = [] + fw = "fw" + with open(infile, "r") as ff: # Read until the line with [Data] for line in ff: header.append(line) fw = line.rsplit() - if (fw[0] == "[Data]"): + if fw[0] == "[Data]": break - + # Read the column names and add the 'State' one - line = ff.readline() - new_head = str(line.rstrip())+', '+state_col+' \n' + line = ff.readline() + new_head = str(line.rstrip()) + ", " + state_col + " \n" header.append(new_head) # Create a temporary file with the new header - tmp_file = 'tmp.csv' ; ihead = 0 - with open(tmp_file, 'w') as tf: + tmp_file = "tmp.csv" + ihead = 0 + with open(tmp_file, "w") as tf: for item in header: tf.write(str(item)) ihead += 1 # The State column - state = [] ; data = [] - last_step = -99 # Create a starting last state value - last_t = 99. # Create a starting step time value + state = [] + data = [] + last_step = -99 # Create a starting last state value + last_t = 99.0 # Create a starting step time value steps = [] - # Read the data adding values to the state - for il,line in enumerate(ff): + # Read the data adding values to the state + for il, line in enumerate(ff): data.append(line) - step = line.split(',')[icol] ; steps.append(step) - stime = float(line.split(',')[icolt]) - if (step == last_step and stime > last_t): + step = line.split(",")[icol] + steps.append(step) + stime = float(line.split(",")[icolt]) + if step == last_step and stime > last_t: state.append(1) else: state.append(0) # Change the previous value - if (len(state)>1): - if(state[-2] == 0): - if (last_t < eps): - # Single measurement + if len(state) > 1: + if state[-2] == 0: + if last_t < eps: + # Single measurement state[-2] = -1 else: # Jump lines affected by software bug and # 2 single measurements in a row # (this can happen when current overshoots) state[-2] = -99 - if (verbose): + if verbose: # Line count starts with 1, thus (il+1) - print('WARNING line=',str(il+ihead),', last step time='+ str(last_t),': Measurement to be ignored') - if (state[-3] == -1): + print( + "WARNING line=", + str(il + ihead), + ", last step time=" + str(last_t), + ": Measurement to be ignored", + ) + if state[-3] == -1: # Avoid 2 single measurements in a row state[-3] = -99 - if (verbose): - print('WARNING Measurement to be ignored: line=',str(il-1+ihead),', last step time='+ str(last_t)) - - elif(state[-2] == 1): + if verbose: + print( + "WARNING Measurement to be ignored: line=", + str(il - 1 + ihead), + ", last step time=" + str(last_t), + ) + + elif state[-2] == 1: state[-2] = 2 else: - sys.exit('STOP function novonix_add_state \n'+\ - 'REASON unexpected state \n'+\ - ' '+str(infile)) + sys.exit( + "STOP function novonix_add_state \n" + + "REASON unexpected state \n" + + " " + + str(infile) + ) last_step = step last_t = stime state[-1] = 2 # Check if the first and last state values have adequate values - if (state[0] != 0 or state[-1] != 2): - sys.exit('STOP function novonix_add_state \n'+\ - 'REASON the State column was not properly populated \n'+\ - ' '+str(infile)) + if state[0] != 0 or state[-1] != 2: + sys.exit( + "STOP function novonix_add_state \n" + + "REASON the State column was not properly populated \n" + + " " + + str(infile) + ) # Check that there are the same numbers of 0s and 2s izeros = np.where(np.array(state) == 0) - itwos = np.where(np.array(state) == 2) - if (np.shape(izeros)[1] != np.shape(itwos)[1]): - sys.exit('STOP function novonix_add_state \n'+\ - 'REASON there is a mismath between State=0 and 2 \n'+\ - ' '+str(infile)) - + itwos = np.where(np.array(state) == 2) + if np.shape(izeros)[1] != np.shape(itwos)[1]: + sys.exit( + "STOP function novonix_add_state \n" + + "REASON there is a mismath between State=0 and 2 \n" + + " " + + str(infile) + ) + # Write the new data to the temporary file - with open(tmp_file, 'a') as tf: - ii = 0 + with open(tmp_file, "a") as tf: + ii = 0 for idata in data: - if (state[ii]>-99): - new_line = str(idata.rstrip())+','+str(state[ii])+'\n' + if state[ii] > -99: + new_line = str(idata.rstrip()) + "," + str(state[ii]) + "\n" tf.write(new_line) - ii+= 1 + ii += 1 # Check the size of the temporal file compared to the original size_original = os.stat(infile).st_size size_tmp = os.stat(tmp_file).st_size - if (size_original > size_tmp): - sys.exit('STOP function novonix_add_state \n'+\ - 'REASON file with State column is smaller than original file \n'+\ - ' '+str(infile)) - + if size_original > size_tmp: + sys.exit( + "STOP function novonix_add_state \n" + + "REASON file with State column is smaller than original file \n" + + " " + + str(infile) + ) + # Replace the input file with the new one - move(tmp_file,infile) + move(tmp_file, infile) if verbose: - print('{} contains now a State column'.format(infile)) - + print("{} contains now a State column".format(infile)) + return -def reduced_protocol(infile,verbose=False): + +def reduced_protocol(infile, verbose=False): """ - Given a Novonix data file, get a reduced protocol + Given a Novonix data file, get a reduced protocol with one command per line. Parameters @@ -610,145 +683,174 @@ def reduced_protocol(infile,verbose=False): """ viable_prot = True - protocol = ['[Reduced Protocol] \n'] - end_rprotocol = '[End Reduced Protocol] \n' + protocol = ["[Reduced Protocol] \n"] + end_rprotocol = "[End Reduced Protocol] \n" # Read the reduced protocol if it is already in place - continue_reading = True ; fw = ' ' - with open(infile,'r') as ff: - while (fw != protocol[0].strip()): - line = ff.readline() ; fw = line.strip() - if (fw == "[Data]"): + continue_reading = True + fw = " " + with open(infile, "r") as ff: + while fw != protocol[0].strip(): + line = ff.readline() + fw = line.strip() + if fw == "[Data]": continue_reading = False break - - if (continue_reading): - while (fw != end_rprotocol.strip()): - line = ff.readline() ; fw = line.strip() - if (fw == "[Data]"): - sys.exit('STOP function reduced_protocol \n'+\ - 'REASON: line [End reduced protocol] not found \n'+\ - ' '+str(infile)+' \n') + + if continue_reading: + while fw != end_rprotocol.strip(): + line = ff.readline() + fw = line.strip() + if fw == "[Data]": + sys.exit( + "STOP function reduced_protocol \n" + + "REASON: line [End reduced protocol] not found \n" + + " " + + str(infile) + + " \n" + ) protocol.append(line) return protocol, viable_prot # Create the reduced protocol (if it does not already exist) - with open(infile,'r') as ff: + with open(infile, "r") as ff: # Read until the protocol starts for line in ff: - if ("[Protocol]" in line): + if "[Protocol]" in line: break # Read until the first character is '[' and compare to the commands continue_reading = True - while (continue_reading): - line = ff.readline() + while continue_reading: + line = ff.readline() - if (line[0] == '['): + if line[0] == "[": # Skip the operating limits if present - if (line.split()[0] != '[Protocol' and - line.split()[0] != '[End'): + if line.split()[0] != "[Protocol" and line.split()[0] != "[End": continue_reading = False # Establish which format is this file - if (':' in line): - fmt_space = False ; commands = com_prot + if ":" in line: + fmt_space = False + commands = com_prot else: fmt_space = True commands = [] for com in com_prot: - newcom = com.replace('_',' ') + newcom = com.replace("_", " ") commands.append(newcom) # Set counters to 0 for the number of lines, steps, # 'State' blocks (0,1,...,1,2), commands and subcommands - iline = 0 ; new_line = ' ' - istep = 0 ; nstep = 0 ; inrepeat = False + iline = 0 + new_line = " " + istep = 0 + nstep = 0 + inrepeat = False istate = 0 - command = ' ' ; isub = 0 + command = " " + isub = 0 # Read until the end of the protocol - while ('[End Protocol]' not in line): + while "[End Protocol]" not in line: fw = line.strip() # Find commands ignoring left spaces - if (fmt_space): + if fmt_space: command = fw[1:-1] else: - try: - command = fw.split(':')[1].strip() - except: + if ":" in fw: + command = fw.split(":")[1].strip() + else: command = fw[1:-1] - if (command in commands or command.split()[0] == 'Repeat'): + if command in commands or command.split()[0] == "Repeat": iline += 1 - + # Add previous line - if (iline>1): + if iline > 1: # Append the new line - protocol.append(new_line+'] \n') - new_line = ' ' + protocol.append(new_line + "] \n") + new_line = " " # Create an 'End Repeat line if needed - if (istep == nstep and inrepeat): - new_line = '['+str(iline)+' : End Repeat '+str(nstep)+' steps :' - protocol.append(new_line+'] \n') - new_line = ' ' - + if istep == nstep and inrepeat: + new_line = ( + "[" + + str(iline) + + " : End Repeat " + + str(nstep) + + " steps :" + ) + protocol.append(new_line + "] \n") + new_line = " " + iline += 1 inrepeat = False # Get the number of repetitions and repeated steps - if (command.split()[0] == 'Repeat'): - if (inrepeat): - sys.exit('STOP function reduced_protocol \n'+\ - 'REASON code not set to handle nested loops \n'+\ - ' '+str(infile)+' \n') + if command.split()[0] == "Repeat": + if inrepeat: + sys.exit( + "STOP function reduced_protocol \n" + + "REASON code not set to handle nested loops \n" + + " " + + str(infile) + + " \n" + ) unexpected = False - if (fmt_space): - line = ff.readline() ; fw = line.strip() - if (fw[1:-1].split()[0] == 'Repeat'): + if fmt_space: + line = ff.readline() + fw = line.strip() + if fw[1:-1].split()[0] == "Repeat": ncount = fw[1:-1].split()[2] else: unexpected = True - line = ff.readline() ; fw = line.strip() - if (fw[1:-1].split()[0] == 'Step'): + line = ff.readline() + fw = line.strip() + if fw[1:-1].split()[0] == "Step": nstep = int(fw[1:-1].split()[2]) else: unexpected = True else: - try: - ncount = fw.split(':')[2].strip().split()[0] - except: + if ":" in fw: + ncount = fw.split(":")[2].strip().split()[0] + else: unexpected = True - try: - nstep = int(fw.split(':')[3][:-1]) - except: + if ":" in fw: + nstep = int(fw.split(":")[3][:-1]) + else: unexpected = True if unexpected: - sys.exit('STOP function reduced_protocol \n'+\ - 'REASON unexpected protocol syntax \n'+\ - ' '+str(infile)+' \n') - - new_line = '['+str(iline)+' : Repeat '+\ - ncount+' times :' - istep = 0 ; inrepeat = True + sys.exit( + "STOP function reduced_protocol \n" + + "REASON unexpected protocol syntax \n" + + " " + + str(infile) + + " \n" + ) + + new_line = "[" + str(iline) + " : Repeat " + ncount + " times :" + istep = 0 + inrepeat = True else: - if (inrepeat): + if inrepeat: istep += 1 istate = istate + int(ncount) else: - istate +=1 - - new_line = '['+str(iline)+' : '+\ - command.replace(' ','_')+' : ' - - elif (inrepeat and - (command == increment1 or - command == increment1.replace(' ','_') or - command == increment2 or - command == increment2.replace(' ','_'))): + istate += 1 + + new_line = ( + "[" + str(iline) + " : " + command.replace(" ", "_") + " : " + ) + + elif inrepeat and ( + command == increment1 + or command == increment1.replace(" ", "_") + or command == increment2 + or command == increment2.replace(" ", "_") + ): # Substract any Increment step nstep = nstep - 1 @@ -756,67 +858,92 @@ def reduced_protocol(infile,verbose=False): # Append subcommands without brackets, # separated by semicoloms isub += 1 - + subcommand = fw[1:-1] - if (subcommand not in ignore): - if (isub == 1): + if subcommand not in ignore: + if isub == 1: new_line = new_line + subcommand else: - new_line = new_line + ';' + subcommand - + new_line = new_line + ";" + subcommand + # Continue reading line = ff.readline() # Add last line if not done already - ilast = int(protocol[-1].split()[0].split('[')[1]) + ilast = int(protocol[-1].split()[0].split("[")[1]) - if (iline>ilast): - protocol.append(new_line+'] \n') + if iline > ilast: + protocol.append(new_line + "] \n") # Create an 'End Repeat line if needed - if (istep == nstep and inrepeat): - new_line = '['+str(iline+1)+' : End Repeat '+str(nstep)+' steps :' - protocol.append(new_line+'] \n') - new_line = ' ' - + if istep == nstep and inrepeat: + new_line = ( + "[" + str(iline + 1) + " : End Repeat " + str(nstep) + " steps :" + ) + protocol.append(new_line + "] \n") + new_line = " " + protocol.append(end_rprotocol) # Test that the number of protocol lines taking into account repetitions. - step_number = read_column(infile,col_step,outtype='int') - state_number = read_column(infile,state_col,outtype='int') + step_number = read_column(infile, col_step, outtype="int") + state_number = read_column(infile, state_col, outtype="int") # Find the number of different steps (CC-CV is considered one) - ind = np.where((state_number < 1) & - (step_number != CCCV_CVc) & - (step_number != CCCV_CVd)) + ind = np.where( + (state_number < 1) & (step_number != CCCV_CVc) & (step_number != CCCV_CVd) + ) uniq_step = np.shape(ind)[1] - minus1 = np.shape(np.where((state_number == -1) & - (step_number != CCCV_CVc) & (step_number != CCCV_CVd)))[1] + minus1 = np.shape( + np.where( + (state_number == -1) + & (step_number != CCCV_CVc) + & (step_number != CCCV_CVd) + ) + )[1] if verbose: - print('Unique steps = {} (step=-1: {}), Steps from protocol = {}'.format(uniq_step,minus1,istate)) - - if(istate > uniq_step and verbose): - print('WARNING function reduced_protocol \n'+\ - 'REASON Mismatched protocol ('+str(istate)+\ - ') and steps ('+str(uniq_step)+'): '+str(infile)+' \n') - if(istate < uniq_step): + print( + "Unique steps = {} (step=-1: {}), Steps from protocol = {}".format( + uniq_step, minus1, istate + ) + ) + + if istate > uniq_step and verbose: + print( + "WARNING function reduced_protocol \n" + + "REASON Mismatched protocol (" + + str(istate) + + ") and steps (" + + str(uniq_step) + + "): " + + str(infile) + + " \n" + ) + if istate < uniq_step: viable_prot = False - if (verbose): - print('WARNING function reduced_protocol \n'+\ - 'REASON Less protocol lines ('+str(istate)+\ - ') than actual steps ('+str(uniq_step)+'): \n'+ - str(infile)+' \n') - - return protocol,viable_prot - -def novonix_add_loopnr(infile,verbose=False): + if verbose: + print( + "WARNING function reduced_protocol \n" + + "REASON Less protocol lines (" + + str(istate) + + ") than actual steps (" + + str(uniq_step) + + "): \n" + + str(infile) + + " \n" + ) + + return protocol, viable_prot + + +def novonix_add_loopnr(infile, verbose=False): """ - Given a cleaned Novonix data file, it adds a 'Loop number' column, - with monotonically increasing numbers and - the protocol line corresponding to a given measurement. + Given a cleaned Novonix data file, it adds a 'Loop number' column, + with monotonically increasing numbers and + the protocol line corresponding to a given measurement. Measurements that are not being repeated are assinged: Loop number=0. - This values are determined by the change in the 'State' and + This values are determined by the change in the 'State' and the protocol description in the header Parameters @@ -840,177 +967,214 @@ def novonix_add_loopnr(infile,verbose=False): """ # Check if the file already has the new State column - icol = icolumn(infile,loop_col) - if (icol>-1): + icol = icolumn(infile, loop_col) + if icol > -1: if verbose: - print('The file {} already has a Loop number column'.format(infile)) + print("The file {} already has a Loop number column".format(infile)) return - + # Get the reduced protocol as a list - protocol, viable_prot = reduced_protocol(infile,verbose=verbose) + protocol, viable_prot = reduced_protocol(infile, verbose=verbose) # Read the Step_Number column - steps = read_column(infile, col_step, outtype='int') + steps = read_column(infile, col_step, outtype="int") # Read the States and the set of measurements for a single state - states = read_column(infile, state_col,outtype='int') - izeros, = np.where(np.logical_or(states == 0,states == -1)) - itwos, = np.where(np.logical_or(states == 2, states == -1)) + states = read_column(infile, state_col, outtype="int") + izeros, = np.where(np.logical_or(states == 0, states == -1)) + itwos, = np.where(np.logical_or(states == 2, states == -1)) - linenr = np.zeros(shape=len(steps),dtype=int) ; linenr.fill(-999.) - loopnr = np.zeros(shape=len(steps),dtype=int) ; loopnr.fill(-999.) + linenr = np.zeros(shape=len(steps), dtype=int) + linenr.fill(-999.0) + loopnr = np.zeros(shape=len(steps), dtype=int) + loopnr.fill(-999.0) - if (viable_prot): + if viable_prot: loopnr.fill(0) - + # Prepare to loop over the protocol lines prot = protocol[1:-1] iprot = 0 - inrepeat = False ; com_repeat = [] - ntimes = 0 ; itimes = 0 ; iloop = 0 + inrepeat = False + com_repeat = [] + ntimes = 0 + itimes = 0 + iloop = 0 last_step = -1 # Loop over the measurement sets - for iz,it in zip(izeros,itwos): + for iz, it in zip(izeros, itwos): # Step of the Data Subset step = steps[iz] # Read protocol line continue_reading = True - while (continue_reading): - command = prot[iprot].split(':')[1].strip() - if(command.split(' ')[0].strip() == 'Repeat'): - inrepeat = True ; com_repeat = [] - iloop += 1 - itimes = 0 ; irstep = 0 - ntimes = int(command.split(' ')[1].strip()) - + while continue_reading: + command = prot[iprot].split(":")[1].strip() + if command.split(" ")[0].strip() == "Repeat": + inrepeat = True + com_repeat = [] + iloop += 1 + itimes = 0 + irstep = 0 + ntimes = int(command.split(" ")[1].strip()) + iprot += 1 - first_rep = iprot + 1 + first_rep = iprot + 1 firstend = True - - elif(command.split(' ')[0].strip() == 'End'): - nrstep = int(command.split(' ')[2].strip()) + + elif command.split(" ")[0].strip() == "End": + nrstep = int(command.split(" ")[2].strip()) if firstend: - irstep = 0 ; itimes += 1 + irstep = 0 + itimes += 1 iloop += 1 firstend = False - + # Check that the array of repeated steps has the expected lenght - if (len(com_repeat) != nrstep): - sys.exit('STOP function novonix_add_loopnr \n'+\ - 'REASON array of repeated steps has an unexpected length \n'+\ - ' '+str(len(com_repeat))+' != '+str(nrstep)+' \n'+ - ' '+str(infile)+' \n') - + if len(com_repeat) != nrstep: + sys.exit( + "STOP function novonix_add_loopnr \n" + + "REASON array of repeated steps has an unexpected length \n" + + " " + + str(len(com_repeat)) + + " != " + + str(nrstep) + + " \n" + + " " + + str(infile) + + " \n" + ) + # Repeated commands index = com_prot.index(com_repeat[irstep]) sel = select_com_val(index) - if(sel): - if ((step==CCCV_CVc or step==CCCV_CVd) and - (last_step==CCCV_CCc or last_step==CCCV_CCd)): + if sel: + if (step == CCCV_CVc or step == CCCV_CVd) and ( + last_step == CCCV_CCc or last_step == CCCV_CCd + ): # 1 protocol line for CC-CV last_step = -1 irstep = irstep - 1 - + # Compare the protocol and step values - linenr[iz:it+1] = first_rep + irstep - loopnr[iz:it+1] = iloop + linenr[iz : it + 1] = first_rep + irstep + loopnr[iz : it + 1] = iloop continue_reading = False - + # Adequately deal with the reduced protocol counters - if (itimes == ntimes-1 and irstep == nrstep-1): + if itimes == ntimes - 1 and irstep == nrstep - 1: # Last command of the last repetition: reset iprot += 1 inrepeat = False - itimes = 0 ; ntimes = 0 - nrstep = 0 ; first_rep = 0 - elif(itimes < ntimes-1 and irstep == nrstep-1): + itimes = 0 + ntimes = 0 + nrstep = 0 + first_rep = 0 + elif itimes < ntimes - 1 and irstep == nrstep - 1: # Last command within a repetition - irstep = 0 ; itimes += 1 - iloop += 1 + irstep = 0 + itimes += 1 + iloop += 1 else: irstep += 1 - - elif(command in com_prot[:-1]): + + elif command in com_prot[:-1]: index = com_prot.index(command) sel = select_com_val(index) - if (sel): - if ((step==CCCV_CVc or step==CCCV_CVd) and - (last_step==CCCV_CCc or last_step==CCCV_CCd)): + if sel: + if (step == CCCV_CVc or step == CCCV_CVd) and ( + last_step == CCCV_CCc or last_step == CCCV_CCd + ): # 1 protocol line for CC-CV last_step = -1 iprot = iprot - 1 else: if inrepeat: com_repeat.append(command) - - linenr[iz:it+1] = iprot + 1 + + linenr[iz : it + 1] = iprot + 1 if inrepeat: - loopnr[iz:it+1] = iloop + loopnr[iz : it + 1] = iloop continue_reading = False - + iprot += 1 - if (iprot == len(prot)): + if iprot == len(prot): continue_reading = False else: - sys.exit('STOP function novonix_add_loopnr \n'+\ - 'REASON unexpected command in reduced protocol \n'+\ - ' '+str(infile)+' \n') + sys.exit( + "STOP function novonix_add_loopnr \n" + + "REASON unexpected command in reduced protocol \n" + + " " + + str(infile) + + " \n" + ) last_step = step - + # Create a temporary file with the new header - header = [] ; fw = 'fw' - tmp_file = 'tmp.csv' - with open(infile,'r') as ff: + header = [] + fw = "fw" + tmp_file = "tmp.csv" + with open(infile, "r") as ff: # Read until the line with [End Protocol] - while (fw != "[Data]"): - line = ff.readline() ; header.append(line) - fw = line.split()[0] + while fw != "[Data]": + line = ff.readline() + header.append(line) + fw = line.split()[0] # Create a temporary file with the header as it was - with open(tmp_file, 'w') as tf: + with open(tmp_file, "w") as tf: for item in header[:-1]: tf.write(str(item)) # Add the reduced protocol - with open(tmp_file, 'a') as tf: + with open(tmp_file, "a") as tf: for item in protocol: tf.write(str(item)) # Add [Data] line tf.write(str(header[-1])) # Read the column names and add the 'Loop' and 'Line' - line = ff.readline() - new_head = str(line.rstrip())+\ - ', '+line_col+', '+loop_col+' \n' - with open(tmp_file, 'a') as tf: + line = ff.readline() + new_head = str(line.rstrip()) + ", " + line_col + ", " + loop_col + " \n" + with open(tmp_file, "a") as tf: tf.write(str(new_head)) # Write the data + 2 new columns in the temporary file - for ii,line in enumerate(ff): + for ii, line in enumerate(ff): # Write the new data to the temporary file - with open(tmp_file, 'a') as tf: - new_line = line.rstrip()+','+\ - str(linenr[ii])+','+str(loopnr[ii])+'\n' + with open(tmp_file, "a") as tf: + new_line = ( + line.rstrip() + "," + str(linenr[ii]) + "," + str(loopnr[ii]) + "\n" + ) tf.write(new_line) # Check the size of the temporal file compared to the original size_original = os.stat(infile).st_size size_tmp = os.stat(tmp_file).st_size - if (size_original > size_tmp): - sys.exit('STOP function novonix_add_loopnr \n'+\ - 'REASON new file is smaller than original file \n'+\ - ' '+str(infile)) - + if size_original > size_tmp: + sys.exit( + "STOP function novonix_add_loopnr \n" + + "REASON new file is smaller than original file \n" + + " " + + str(infile) + ) + # Replace the input file with the new one - move(tmp_file,infile) + move(tmp_file, infile) if verbose: - print('{} contains now the columns Loop number and Protocol line'.format(infile)) + print( + "{} contains now the columns Loop number and Protocol line".format( + infile + ) + ) return -def prepare_novonix(infile,addstate=False,lprotocol=False,overwrite=False,verbose=False): + +def prepare_novonix( + infile, addstate=False, lprotocol=False, overwrite=False, verbose=False +): """ Given a Novonix data file, it prepare it to be handled. @@ -1046,38 +1210,42 @@ def prepare_novonix(infile,addstate=False,lprotocol=False,overwrite=False,verbos >>>> prep.prepare_novonix('example_data/example_data.csv',addstate=True,lprotocol=True,overwrite=False,verbose= False) File example_data_prep.csv has been prepared. """ - + # Extract the path and file name dirname, fname = os.path.split(os.path.abspath(infile)) - if (not overwrite): - root = fname.split('.')[0] - ending = fname.split('.')[1] - fname = root+'_prep.'+ending - # If *prep* file already exists, it will be replaced. - copy(infile,os.path.join(dirname,fname)) - + if not overwrite: + root = fname.split(".")[0] + ending = fname.split(".")[1] + fname = root + "_prep." + ending + # If *prep* file already exists, it will be replaced. + copy(infile, os.path.join(dirname, fname)) + # Modify the slashes in the input path if needed - file_to_open = Path(dirname) / fname + file_to_open = Path(dirname) / fname # Check if the file has the expected structure for a Novonix file answer = isnovonix(file_to_open) - if (not answer): sys.exit('STOP Input not from Novonix, {}'.format(file_to_open)) + if not answer: + sys.exit("STOP Input not from Novonix, {}".format(file_to_open)) # Clean the Novonix file cleannovonix(file_to_open) if addstate: # Check if the file has a State column and if not, create it - novonix_add_state(file_to_open,verbose=verbose) + novonix_add_state(file_to_open, verbose=verbose) if lprotocol: # Check if the file has a Loop number and Protocol line columns # and if not, create it - novonix_add_loopnr(file_to_open,verbose=verbose) + novonix_add_loopnr(file_to_open, verbose=verbose) + + print("File {} has been prepared.".format(fname)) + - print('File {} has been prepared.'.format(fname)) - -if __name__ == '__main__': +if __name__ == "__main__": if len(sys.argv) > 1: - prepare_novonix(sys.argv[1],addstate=True,lprotocol=True,overwrite=False,verbose=True) + prepare_novonix( + sys.argv[1], addstate=True, lprotocol=True, overwrite=False, verbose=True + ) diff --git a/preparenovonix/test_novonix_add.py b/preparenovonix/test_novonix_add.py index 7b8aa5f..83b8d50 100644 --- a/preparenovonix/test_novonix_add.py +++ b/preparenovonix/test_novonix_add.py @@ -1,32 +1,35 @@ import preparenovonix.novonix_add as prep -import numpy as np -import pytest -exfile = 'example_data/example_data.csv' -exfile_prep = 'example_data/example_data_prep.csv' +exfile = "example_data/example_data.csv" +exfile_prep = "example_data/example_data_prep.csv" + def test_select_com_val(): - one = 'step == com_val1[index]' - two = 'np.logical_or(step == com_val1[index], step == com_val2[index])' + one = "step == com_val1[index]" + two = "np.logical_or(step == com_val1[index], step == com_val2[index])" assert prep.select_com_val(0) == one assert prep.select_com_val(1) == one assert prep.select_com_val(2) == two assert prep.select_com_val(3) == two assert prep.select_com_val(4) == two + def test_isnovonix(): - assert prep.isnovonix(exfile) == True - assert prep.isnovonix('novonix_add') == False + assert prep.isnovonix(exfile) is True + assert prep.isnovonix("novonix_add") is False + def test_icolumn(): - assert prep.icolumn(exfile_prep,'Step Number') > -1 - assert prep.icolumn(exfile_prep,'Step Time (h)') > -1 + assert prep.icolumn(exfile_prep, "Step Number") > -1 + assert prep.icolumn(exfile_prep, "Step Time (h)") > -1 + def test_read_column(): - col = prep.read_column(exfile_prep,'Step Number',outtype='int') + col = prep.read_column(exfile_prep, "Step Number", outtype="int") assert len(col) > 1 assert min(col) > -1 + def test_reduced_protocol(): prot = prep.reduced_protocol(exfile_prep) assert len(prot) > 1 diff --git a/pyproject.toml b/pyproject.toml deleted file mode 100644 index dfab025..0000000 --- a/pyproject.toml +++ /dev/null @@ -1,16 +0,0 @@ -[tool.black] -line-length = 79 -include = '\.pyi?$' -exclude = ''' -/( - \.git - | \.hg - | \.mypy_cache - | \.tox - | \.venv - | _build - | buck-out - | build - | dist -)/ -'''