simulation_parameters_ESS.py

import json
import numpy as np
import numpy.random as rnd
import os
import utils

class parameter_storage(object):
    """
    This class contains the simulation parameters in a dictionary called params.
    """

    def __init__(self, fn=None):

        if fn == None:
            self.params = {}
            self.set_default_params()
            self.set_filenames()
        else:
            self.params = self.load_params_from_file(fn)

    def set_default_params(self):
        self.params['simulator'] = 'hw_ess'
        self.params['hardware_setup'] = 'medium'
        """
        hardwareSetup   #HICANNs    #Neurons
        'one-hicann'    1   59
        'one-reticle'   8 (4x2)     472
        'small'     32( 8x4)    1888
        'medium'    128 (16x8)  7552
        'medium2'   128 (32x4)  7552
        'large'     240 (20x12)     14160
        'large2'    224 (28x8)  13216
        'one-wafer'     384     22656
        """

        # new parameter make distinguishment between one / two dimensional model easy
        self.params['n_grid_dimensions'] = 1
        self.params['ess_speedup'] = 100000

        # ###################
        # HEXGRID PARAMETERS
        # ###################
        self.params['N_RF'] = 50# for more than 2-D tuning space: np.int(n_cells/N_V/N_theta)
        self.params['N_RF_X'] = self.params['N_RF']
        self.params['N_RF_Y'] = 1
        self.params['N_V'], self.params['N_theta'] = 10, 1# resolution in velocity norm and direction
        self.params['n_orientation'] = 1 # only for compatitibility with code borrowed from  v1-anticipation branch
        if self.params['n_grid_dimensions'] == 2:
            self.params['n_rf_x'] = np.int(np.sqrt(self.params['n_rf'] * np.sqrt(3)))
            self.params['n_rf_y'] = np.int(np.sqrt(self.params['n_rf']))
            # np.sqrt(np.sqrt(3)) comes from resolving the problem "how to quantize the square with a hex grid of a total of n_rfdots?"
            self.params['n_theta'] = 1# resolution in velocity norm and direction
        else:
            self.params['N_RF_X'] = self.params['N_RF']
            self.params['N_RF_Y'] = 1
            self.params['N_theta'] = 1
        self.params['log_scale'] = 2.0 # base of the logarithmic tiling of particle_grid; linear if equal to one
        self.params['sigma_RF_pos'] = .05 # some variability in the position of RFs
        self.params['sigma_RF_speed'] = .30 # some variability in the speed of RFs
        self.params['sigma_RF_direction'] = .25 * 2 * np.pi # some variability in the direction of RFs
        self.params['sigma_RF_orientation'] = .1 * np.pi

        # ###################
        # NETWORK PARAMETERS
        # ###################
        self.params['n_exc'] = self.params['N_RF_X'] * self.params['N_RF_Y'] * self.params['N_V'] * self.params['N_theta'] # number of excitatory cells per minicolumn
        self.params['n_exc_per_mc'] = 1             # only for compatitibility with code borrowed from  v1-anticipation branch
        self.params['fraction_inh_cells'] = 0.20    # fraction of inhibitory cells in the network, only approximately!
        # for 2-Dimensional tuning property space use sqrt(fraction_inh_cells), in n-dim tuning-property space use fraction_inh_cells**(1./n_dim)
        self.params['N_V_INH'] = int(round(np.sqrt(self.params['fraction_inh_cells']) * self.params['N_V']))
        self.params['N_X_INH'] = int(round(np.sqrt(self.params['fraction_inh_cells']) * self.params['N_RF_X']))
        self.params['N_Y_INH'] = 1               # for 2-dim tuning space only
        self.params['N_theta_inh'] = 1
        self.params['N_RF_INH'] = self.params['N_V_INH'] * self.params['N_X_INH'] * self.params['N_theta_inh']
        self.params['n_inh' ] = self.params['N_RF_INH']
        self.params['n_cells'] = self.params['n_exc'] + self.params['n_inh']

        # ###################
        # CELL PARAMETERS   #
        # ###################
        self.params['neuron_model'] = 'IF_cond_exp'
#        self.params['neuron_model'] = 'IF_cond_alpha'
#        self.params['neuron_model'] = 'EIF_cond_exp_isfa_ista'
        self.params['tau_syn_exc'] = 5.0 # 10.
        self.params['tau_syn_inh'] = 5.0 # 20.
        if self.params['neuron_model'] == 'IF_cond_exp':
            self.params['cell_params_exc'] = {'cm':0.2, 'tau_refrac':1.0, 'v_thresh':-30.0, 'tau_syn_E': self.params['tau_syn_exc'], 'tau_syn_I':self.params['tau_syn_inh'], 'tau_m' : 90., 'v_reset' : -70., 'v_rest':-50, 'e_rev_I':-100.}
            self.params['cell_params_inh'] = {'cm':0.2, 'tau_refrac':1.0, 'v_thresh':-30.0, 'tau_syn_E': self.params['tau_syn_exc'], 'tau_syn_I':self.params['tau_syn_inh'], 'tau_m' : 90., 'v_reset' : -70., 'v_rest':-50, 'e_rev_I':-100.}
        elif self.params['neuron_model'] == 'EIF_cond_exp_isfa_ista':
            self.params['cell_params_exc'] = {'cm':0.2, 'tau_refrac':1.0, 'v_thresh':-30.0, 'tau_syn_E':self.params['tau_syn_exc'], 'tau_syn_I':self.params['tau_syn_inh'], 'tau_m' : 90., 'v_reset' : -70., 'v_rest':-50., \
                    'b' : 0.5, 'a':4., 'e_rev_I':-100.}
            self.params['cell_params_inh'] = {'cm':0.2, 'tau_refrac':1.0, 'v_thresh':-30.0, 'tau_syn_E':self.params['tau_syn_exc'], 'tau_syn_I':self.params['tau_syn_inh'], 'tau_m' : 90., 'v_reset' : -70., 'v_rest':-50., \
                    'b' : 0.5, 'a':4., 'e_rev_I':-100.}
        # default parameters: /usr/local/lib/python2.6/dist-packages/pyNN/standardmodels/cells.py
        # v = voltage
        self.params['v_init'] = -65.                 # [mV]
        self.params['v_init_sigma'] = 10.             # [mV]

        # v = speed preference
        self.params['v_max_tp'] = 3.0   # [Hz] maximal velocity in visual space for tuning proprties (for each component), 1. means the whole visual field is traversed within 1 second
        self.params['v_min_tp'] = 0.05  # [a.u.] minimal velocity in visual space for tuning property distribution


        # #######################
        # CONNECTIVITY PARAMETERS
        # #######################
        """
        For each connection type ('ee', 'ei', 'ie', 'ii') choose one form of connectivity
        """
        self.params['connectivity_ee'] = 'anisotropic'
        #self.params['connectivity_ee'] = 'isotropic'
#        self.params['connectivity_ee'] = 'random'
#        self.params['connectivity_ee'] = False
#        self.params['connectivity_ei'] = 'anisotropic'
        self.params['connectivity_ei'] = 'isotropic'
#        self.params['connectivity_ei'] = 'random'
#        self.params['connectivity_ei'] = False
#        self.params['connectivity_ie'] = 'anisotropic'
        self.params['connectivity_ie'] = 'isotropic'
#        self.params['connectivity_ie'] = 'random'
#        self.params['connectivity_ie'] = False
#        self.params['connectivity_ii'] = 'anisotropic'
        self.params['connectivity_ii'] = 'isotropic'
#        self.params['connectivity_ii'] = 'random'
#        self.params['connectivity_ii'] = False

        self.params['conn_conf'] = 'motion-based'
#        self.params['conn_conf'] = 'direction-based'
#        self.params['conn_conf'] = 'orientation-direction'
        # when the initial connections are derived on the cell's tuning properties, these two values are used
        self.params['connectivity_radius'] = 1.00      # this determines how much the directional tuning of the src is considered when drawing connections, the connectivity_radius affects the choice w_sigma_x/v
        self.params['delay_scale'] = 1000.      # this determines the scaling
        #from the latency in second (d(src, tgt) / v_src)  to the connection
        #delay (delay_ij = latency_ij * delay_scale) in ms
        self.params['delay_range'] = (0.1, 2.) # [ms], restricts remaining connections to have delays within this range
        self.params['all_connections_have_equal_delays'] = True # if True: IJCNN-paper like motion-based-prediction/anticipation, all have the same delay
        self.params['tau_prediction'] = .001 # [s] fixed latency for neural signaling, determines preferred projection sites of neurons

        self.params['w_sigma_x'] = 0.1 # width of connectivity profile for pre-computed weights
        self.params['w_sigma_v'] = 0.1 # small w_sigma: tuning_properties get stronger weight when deciding on connection
                                       # large w_sigma: make conn prob independent of tuning_properties
#        self.params['w_sigma_v'] = 3.0 # small w_sigma: tuning_properties get stronger weight when deciding on connection
        self.params['w_sigma_isotropic'] = 0.20 # spatial reach of isotropic connectivity, should not be below 0.05 otherwise you don't get the desired p_effective

#        self.params['equal_weights'] = True # if True, connection weights are all equal and w_sigma_ determine only connection probability
        self.params['equal_weights'] = False # if True, connection weights are all equal and w_sigma_ determine only connection probability

        # for anisotropic connections each target cell receives a defined sum of incoming connection weights
        self.params['w_tgt_in_per_cell_ee'] = 0.30 # [uS] how much input should an exc cell get from its exc source cells?
        self.params['w_tgt_in_per_cell_ei'] = 0.45 # [uS] how much input should an inh cell get from its exc source cells?
        self.params['w_tgt_in_per_cell_ie'] = 1.60 # [uS] how much input should an exc cell get from its inh source cells?
        self.params['w_tgt_in_per_cell_ii'] = 0.15 # [uS] how much input should an inh cell get from its source cells?

        #self.params['w_tgt_in_per_cell_ee'] = 0.30 # [uS] how much input should an exc cell get from its exc source cells?
        #self.params['w_tgt_in_per_cell_ei'] = 1. * self.params['w_tgt_in_per_cell_ee']
        #self.params['w_tgt_in_per_cell_ie'] = 1. * self.params['w_tgt_in_per_cell_ee'] / self.params['fraction_inh_cells']

        #self.params['w_tgt_in_per_cell_ii'] = 0.10 # [uS] how much input should an inh cell get from its source cells?
        self.params['w_tgt_in_per_cell_ee'] *= 5. / self.params['tau_syn_exc']
        self.params['w_tgt_in_per_cell_ei'] *= 5. / self.params['tau_syn_exc']
        self.params['w_tgt_in_per_cell_ie'] *= 5. / self.params['tau_syn_inh']
        self.params['w_tgt_in_per_cell_ii'] *= 5. / self.params['tau_syn_inh']
        self.params['w_sigma_distribution'] = 0.2 # percentage of w_mean_isotropic for the sigma of the weight distribution (gaussian) when drawn for isotropic connectivity
        self.params['conn_types'] = ['ee', 'ei', 'ie', 'ii']

        self.params['w_range_hw'] = (0.004516379807692308, 0.04589923076923077)
#        self.params['p_to_w'] =
        self.params['p_ee'] = 0.01              # fraction of network cells allowed to connect to each target cell, used in CreateConnections
        self.params['w_min'] = 5e-4             # When probabilities are transformed to weights, they are scaled so that the map into this range
        self.params['w_max'] = 8e-3
        self.params['n_src_cells_per_neuron'] = round(self.params['p_ee'] * self.params['n_exc']) # only excitatory sources

        # exc - inh
        self.params['p_ei'] = 0.01 #self.params['p_ee']
        self.params['w_ei_mean'] = 0.005
        self.params['w_ei_sigma'] = 0.001

        # inh - exc
        self.params['p_ie'] = 0.01 #self.params['p_ee']
        self.params['w_ie_mean'] = 0.005
        self.params['w_ie_sigma'] = 0.001

        # inh - inh
        self.params['p_ii'] = 0.01
        self.params['w_ii_mean'] = 0.003
        self.params['w_ii_sigma'] = 0.001

        # for random connections only:
        self.params['standard_delay'] = 3           # [ms]
        self.params['standard_delay_sigma'] = 1           # [ms]

        # ######################
        # SIMULATION PARAMETERS
        # ######################
        self.params['seed'] = 0
        self.params['np_random_seed'] = self.params['seed']
        self.params['t_sim'] = 1600.            # [ms] total simulation time
        self.params['t_stimulus'] = 1000.       # [ms] time for a stimulus of speed 1.0 to cross the whole visual field from 0 to 1.
        self.params['t_start'] = 200.           # [ms] Time before stimulus starts
        self.params['t_blank'] = 400.           # [ms] duration of 'blanked' input (if zero, assumes no blank)
        self.params['t_before_blank'] = self.params['t_start'] + 400. # [ms] time when blanking starts, i.e. t_reappear = t_before_blank + t_blank
        self.params['tuning_prop_seed'] = self.params['seed']     # seed for randomized tuning properties
        self.params['input_spikes_seed'] = self.params['seed']
        self.params['dt_sim'] = self.params['delay_range'][0] * 1 # [ms] time step for simulation
        self.params['dt_rate'] = .1             # [ms] time step for the non-homogenous Poisson process
        self.params['n_gids_to_record'] = 30
        self.params['sensory_delay'] = 0.10 # [s] real, physical delay accumulated along the thalamo-cortical pathways
        self.params['compensated_delay'] = 0.10 # [s] assumed sensory delay

        # ######
        # INPUT
        # ######
        self.params['f_max_stim'] = 1000.       # [Hz]
        self.params['w_input_exc'] = 5.00e-2 * 5. / self.params['tau_syn_exc']    # [uS] mean value for input stimulus ---< exc_units (columns
        #self.params['w_input_exc'] = 6.00e-3 * 10. / self.params['tau_syn_exc']    # [uS] mean value for input stimulus ---< exc_units (columns

        # ###############
        # MOTION STIMULUS
        # ###############
        """
        x0 (y0) : start position on x-axis (y-axis)
        u0 (v0) : velocity in x-direction (y-direction)
        """
        self.params['torus_width'] = 1.
        self.params['torus_height'] = 1.
        self.params['motion_params'] = (0.1, 0.5 , 0.3, 0) # stimulus start parameters (x, y, v_x, v_y)
        self.params['blur_X'], self.params['blur_V'] = .05, .05

        self.params['anticipatory_mode'] = True # if True record selected cells to gids_to_record_fn
        self.params['n_cells_to_record_per_location'] = 10
        # record the membrane potentials of some selected cells around these locations:
#        self.params['locations_to_record'] = (.15, .20, .25, .30, .35, .45) # these values + motion_params[0] will determine the cells to record from
        self.params['locations_to_record'] = [  .05 + self.params['motion_params'][0], \
                                                .15 + self.params['motion_params'][0], \
                                                .25 + self.params['motion_params'][0], \
                                                .35 + self.params['motion_params'][0]]

        # the blur parameter represents the input selectivity:
        # high blur means many cells respond to the stimulus
        # low blur means high input selectivity, few cells respond
        # the maximum number of spikes as response to the input alone is not much affected by the blur parameter
        self.params['rf_size_x_gradient'] = .2  # receptive field size for x-pos increases with distance to .5
        self.params['rf_size_y_gradient'] = .2  # receptive field size for y-pos increases with distance to .5
        self.params['rf_size_x_min'] = .02      # cells situated at .5 have this receptive field size
        self.params['rf_size_y_min'] = .02      # cells situated at .5 have this receptive field size
        self.params['rf_size_vx_gradient'] = .2 # receptive field size for vx-pos increases with distance to 0.0
        self.params['rf_size_vy_gradient'] = .2 #
        self.params['rf_size_vx_min'] = .02 # cells situated at .5 have this receptive field size
        self.params['rf_size_vy_min'] = .02 # cells situated at .5 have this receptive field size


        # ######
        # NOISE
        # ######
        self.params['w_exc_noise'] = 4e-3 * 5. / self.params['tau_syn_exc']         # [uS] mean value for noise ---< columns
        self.params['f_exc_noise'] = 2000# [Hz]
        self.params['w_inh_noise'] = 4e-3 * 10. / self.params['tau_syn_inh']         # [uS] mean value for noise ---< columns
        self.params['f_inh_noise'] = 2000# [Hz]

        # no noise:
#        self.params['w_exc_noise'] = 1e-5          # [uS] mean value for noise ---< columns
#        self.params['f_exc_noise'] = 1# [Hz]
#        self.params['w_inh_noise'] = 1e-5          # [uS] mean value for noise ---< columns
#        self.params['f_inh_noise'] = 1# [Hz]
        self.print_cell_numbers()

    def set_folder_name(self, folder_name=None):
        # folder naming code:
        #   PREFIX + XXXX + parameters
        #  X = ['A', # for anisotropic connections
        #       'I', # for isotropic connections
        #       'R', # for random connections
        #       '-', # for non-existant connections
        # order of X: 'ee', 'ei', 'ie', 'ii'

        connectivity_code = ''
        if self.params['connectivity_ee'] == 'anisotropic':
            connectivity_code += 'A'
        elif self.params['connectivity_ee'] == 'isotropic':
            connectivity_code += 'I'
        elif self.params['connectivity_ee'] == 'random':
            connectivity_code += 'R'
        elif self.params['connectivity_ee'] == False:
            connectivity_code += '-'

        if self.params['connectivity_ei'] == 'anisotropic':
            connectivity_code += 'A'
        elif self.params['connectivity_ei'] == 'isotropic':
            connectivity_code += 'I'
        elif self.params['connectivity_ei'] == 'random':
            connectivity_code += 'R'
        elif self.params['connectivity_ei'] == False:
            connectivity_code += '-'

        if self.params['connectivity_ie'] == 'anisotropic':
            connectivity_code += 'A'
        elif self.params['connectivity_ie'] == 'isotropic':
            connectivity_code += 'I'
        elif self.params['connectivity_ie'] == 'random':
            connectivity_code += 'R'
        elif self.params['connectivity_ie'] == False:
            connectivity_code += '-'

        if self.params['connectivity_ii'] == 'anisotropic':
            connectivity_code += 'A'
        elif self.params['connectivity_ii'] == 'isotropic':
            connectivity_code += 'I'
        elif self.params['connectivity_ii'] == 'random':
            connectivity_code += 'R'
        elif self.params['connectivity_ii'] == False:
            connectivity_code += '-'
        self.params['connectivity_code'] = connectivity_code

        if folder_name == None:
            if self.params['neuron_model'] == 'EIF_cond_exp_isfa_ista':
                folder_name = 'AdEx_a%.2e_b%.2e_' % (self.params['cell_params_exc']['a'], self.params['cell_params_exc']['b'])
            else:
               folder_name = 'ESS-WithBlank_delay_%d_%s_nRF%d_tauPred%d_nD%d_delayMax%d_pee%.2e_wee%.2e_wsx%.2e_wsv%.2e_wiso%.2f_taue%d_taui%d_seed%d/' % (\
                       self.params['equal_weights'], self.params['connectivity_code'], self.params['N_RF'], \
                       self.params['tau_prediction'] * 1000., self.params['sensory_delay'] * 1000., \
                       self.params['delay_range'][1], self.params['p_ee'], self.params['w_tgt_in_per_cell_ee'], \
                       self.params['w_sigma_x'], self.params['w_sigma_v'], self.params['w_sigma_isotropic'], \
                       self.params['tau_syn_exc'], self.params['tau_syn_inh'], self.params['seed'])

#            folder_name += connectivity_code
#            folder_name += '/'
            # if parameters should be stored in the folder name:
#            folder_name += "_pee%.1e_wen%.1e_tausynE%d_I%d_bx%.1e_bv%.1e_wsigmax%.2e_wsigmav%.2e_wee%.2e_wei%.2e_wie%.2e_wii%.2e_delay%d_connRadius%.2f/" % \
#                        (self.params['p_ee'], self.params['w_exc_noise'], self.params['tau_syn_exc'], self.params['tau_syn_inh'], self.params['blur_X'], self.params['blur_V'], self.params['w_sigma_x'], self.params['w_sigma_v'], self.params['w_tgt_in_per_cell_ee'], \
#                        self.params['w_tgt_in_per_cell_ei'], self.params['w_tgt_in_per_cell_ie'], self.params['w_tgt_in_per_cell_ii'], self.params['delay_scale'], self.params['connectivity_radius'])

            self.params['folder_name'] = folder_name
        else:
            self.params['folder_name'] = folder_name
        print 'Folder name:', self.params['folder_name']


    def set_filenames(self, folder_name=None):

        self.set_folder_name(folder_name)
        print 'Folder name:', self.params['folder_name']

        # in order to NOT re-compute the input spike trains when the stimulus parameters have not changed, do NOT store them in a subfolder of self.params['folder_name']
        self.params['input_folder'] = "InputSpikeTrains_x0%.2e_v0%.2e_bX%.2e_bV%.2e_fstim%.1e_tsim%d_tblank%d_tbeforeblank%d_%dnrns/" % \
                (self.params['motion_params'][0], self.params['motion_params'][2], self.params['blur_X'], self.params['blur_V'], \
                self.params['f_max_stim'], self.params['t_sim'], self.params['t_blank'], self.params['t_before_blank'], self.params['n_cells'])
        # if you want to store the input files in a subfolder of self.params['folder_name'], do this:
#        self.params['input_folder'] = "%sInputSpikeTrains/"   % self.params['folder_name']# folder containing the input spike trains for the network generated from a certain stimulus
        self.params['spiketimes_folder'] = "%sSpikes/" % self.params['folder_name']
        self.params['volt_folder'] = "%sVoltageTraces/" % self.params['folder_name']
        self.params['parameters_folder'] = "%sParameters/" % self.params['folder_name']
        self.params['connections_folder'] = "%sConnections/" % self.params['folder_name']
        self.params['figures_folder'] = "%sFigures/" % self.params['folder_name']
        self.params['movie_folder'] = "%sMovies/" % self.params['folder_name']
        self.params['tmp_folder'] = "%stmp/" % self.params['folder_name']
        self.params['data_folder'] = '%sData/' % (self.params['folder_name']) # for storage of analysis results etc
        # all folders to be created if not yet existing:
        self.params['folder_names'] = [self.params['folder_name'], \
                            self.params['spiketimes_folder'], \
                            self.params['volt_folder'], \
                            self.params['parameters_folder'], \
                            self.params['connections_folder'], \
                            self.params['figures_folder'], \
                            self.params['movie_folder'], \
                            self.params['tmp_folder'], \
                            self.params['data_folder'], \
                            self.params['input_folder']] 

        self.params['params_fn_json'] = '%ssimulation_parameters.json' % (self.params['parameters_folder'])

        # ESS specific files
        self.params['realized_conn_file'] = '%srealized_connections.txt' % (self.params['connections_folder'])
        self.params['lost_conn_file'] = '%slost_connections.txt' % (self.params['connections_folder'])

        # input spiketrains
        self.params['merged_input_spiketrains_fn'] = "%sinput_spiketrain_merged.dat" % (self.params['input_folder'])
        self.params['input_st_fn_base'] = "%sstim_spike_train_" % self.params['input_folder']# input spike trains filename base
        self.params['input_rate_fn_base'] = "%srate_" % self.params['input_folder']# input spike trains filename base

        # output spiketrains
        self.params['exc_spiketimes_fn_base'] = '%sexc_spikes_' % self.params['spiketimes_folder']
        self.params['exc_spiketimes_fn_merged'] = '%sexc_spikes_merged_' % self.params['spiketimes_folder']
        self.params['exc_nspikes_fn_merged'] = '%sexc_nspikes' % self.params['spiketimes_folder']
        self.params['exc_nspikes_nonzero_fn'] = '%sexc_nspikes_nonzero.dat' % self.params['spiketimes_folder']
        self.params['inh_spiketimes_fn_base'] = '%sinh_spikes_' % self.params['spiketimes_folder']
        self.params['inh_spiketimes_fn_merged'] = '%sinh_spikes_merged_' % self.params['spiketimes_folder']
        self.params['inh_nspikes_fn_merged'] = '%sinh_nspikes' % self.params['spiketimes_folder']
        self.params['inh_nspikes_nonzero_fn'] = '%sinh_nspikes_nonzero.dat' % self.params['spiketimes_folder']
        self.params['exc_volt_fn_base'] = '%sexc_volt' % self.params['volt_folder']
        self.params['exc_gsyn_fn_base'] = '%sexc_gsyn' % self.params['volt_folder']
        self.params['inh_volt_fn_base'] = '%sinh_volt' % self.params['volt_folder']
        self.params['rasterplot_exc_fig'] = '%srasterplot_exc.png' % (self.params['figures_folder'])
        self.params['rasterplot_inh_fig'] = '%srasterplot_inh.png' % (self.params['figures_folder'])

        # tuning properties and other cell parameter files
        self.params['tuning_prop_means_fn'] = '%stuning_prop_means.prm' % (self.params['parameters_folder']) # for excitatory cells
        self.params['tuning_prop_inh_fn'] = '%stuning_prop_inh.prm' % (self.params['parameters_folder']) # for inhibitory cells
        self.params['tuning_prop_fig_exc_fn'] = '%stuning_properties_exc.png' % (self.params['figures_folder'])
        self.params['tuning_prop_fig_inh_fn'] = '%stuning_properties_inh.png' % (self.params['figures_folder'])
        self.params['gids_to_record_fn'] = '%sgids_to_record.dat' % (self.params['parameters_folder'])

        self.params['prediction_fig_fn_base'] = '%sprediction_' % (self.params['figures_folder'])

        # CONNECTION FILES
        self.params['weight_and_delay_fig'] = '%sweights_and_delays.png' % (self.params['figures_folder'])

        # connection lists have the following format: src_gid  tgt_gid  weight  delay
        # E - E
        self.params['conn_list_ee_fn_base'] = '%sconn_list_ee_' % (self.params['connections_folder'])
        self.params['merged_conn_list_ee'] = '%smerged_conn_list_ee.dat' % (self.params['connections_folder'])
        # E - I
        self.params['conn_list_ei_fn_base'] = '%sconn_list_ei_' % (self.params['connections_folder'])
        self.params['merged_conn_list_ei'] = '%smerged_conn_list_ei.dat' % (self.params['connections_folder'])
        # I - E
        self.params['conn_list_ie_fn_base'] = '%sconn_list_ie_' % (self.params['connections_folder'])
        self.params['merged_conn_list_ie'] = '%smerged_conn_list_ie.dat' % (self.params['connections_folder'])
        # I - I
        self.params['conn_list_ii_fn_base'] = '%sconn_list_ii_' % (self.params['connections_folder'])
        self.params['merged_conn_list_ii'] = '%smerged_conn_list_ii.dat' % (self.params['connections_folder'])

        # used for different projections ['ee', 'ei', 'ie', 'ii'] for plotting
        self.params['conn_mat_fn_base'] = '%sconn_mat_' % (self.params['connections_folder'])
        self.params['delay_mat_fn_base'] = '%sdelay_mat_' % (self.params['connections_folder'])

        # ANALYSIS RESULTS
        # these files receive the output folder when they are create / processed --> more suitable for parameter sweeps
        self.params['xdiff_vs_time_fn'] = 'xdiff_vs_time.dat'
        self.params['vdiff_vs_time_fn'] = 'vdiff_vs_time.dat'

    def check_folders(self):
        """
        Returns True if all folders exist, False otherwise
        """
        all_folders_exist = True
        for f in self.params['folder_names']:
            if not os.path.exists(f):
                all_folders_exist = False

        return all_folders_exist

    def create_folders(self, p=None):
        """
        Must be called from 'outside' this class before the simulation
        """
        if p == None:
            p = self.params

        for f in p['folder_names']:
            if not os.path.exists(f):
                print 'Creating folder:\t%s' % f
                os.system("mkdir %s" % (f))

    def load_params(self):
        """
        return the simulation parameters in a dictionary
        """
#        self.ParamSet = ntp.ParameterSet(self.params)
#        return self.ParamSet
        return self.params


    def load_params_from_file(self, fn):

        f = file(fn, 'r')
        print 'Loading parameters from', fn
        self.params = json.load(f)
        return self.params


    def update_values(self, kwargs):
        for key, value in kwargs.iteritems():
            self.params[key] = value
        # update the dependent parameters
        self.set_filenames()
#        self.ParamSet = ntp.ParameterSet(self.params)


    def write_parameters_to_file(self, fn=None, params=None):
        """
        Keyword arguments:
        fn -- (optional) target output filename for json dictionary
        params -- (optional) the modified parameter dictionary that is to write
        """

        if fn == None:
            fn = self.params['params_fn_json']
        if params == None:
            params_to_write = self.params
        if not (os.path.isdir(params_to_write['folder_name'])):
            print 'Creating folder:\n\t%s' % params_to_write['folder_name']
            self.create_folders(params_to_write)
        print 'Writing parameters to: %s' % (fn)
        output_file = file(fn, 'w')
        d = json.dump(params_to_write, output_file, sort_keys=True, indent=4)


    def print_cell_numbers(self):
        print 'Excitatory cells:'
        print 'N_RF_X %d\tN_RF_Y %d\tN_V %d' % (self.params['N_RF_X'], self.params['N_RF_Y'], self.params['N_V'])
        print 'Inhibitory cells:'
        print 'N_X_INH % d\tN_Y_INH %d\tN_V_INH %d' % (self.params['N_X_INH'], self.params['N_Y_INH'], self.params['N_V_INH'])
#        print 'N_HC: %d   N_MC_PER_HC: %d' % (self.params['N_RF_X'] * self.params['N_RF_Y'], self.params['N_V'] * self.params['N_theta'])
        print 'n_cells: %d\tn_exc: %d\tn_inh: %d\nn_inh / n_exc = %.3f\tn_inh / n_cells = %.3f' % (self.params['n_cells'], self.params['n_exc'], self.params['n_inh'], \
                self.params['n_inh'] / float(self.params['n_exc']), self.params['n_inh'] / float(self.params['n_cells']))


class ParameterContainer(parameter_storage):

    def __init__(self, fn):
        super(ParameterContainer, self).__init__()
        self.root_dir = os.path.dirname(fn)
        # If the folder has been moved, all filenames need to be updated
        self.update_values({self.params['folder_name'] : self.root_dir})

    def load_params(self, fn):

        f = file(fn, 'r')
        print 'Loading parameters from', fn
        self.params = json.load(f)

    def update_values(self, kwargs):
        for key, value in kwargs.iteritems():
            self.params[key] = value

        # update the dependent parameters
        # --> to be implemented by another function (e.g. set_filenames())

    def create_folders(self):
        """
        Must be called from 'outside' this class before the simulation
        """
        for f in self.params['folder_names']:
            if not os.path.exists(f):
                print 'Creating folder:\t%s' % f
                os.system("mkdir %s" % (f))

    def load_params(self):
        """
        return the simulation parameters in a dictionary
        """
        return self.ParamSet


    def write_parameters_to_file(self, fn=None):
        if fn == None:
            fn = self.params['params_fn_json']
        print 'Writing parameters to: %s' % (fn)
        self.ParamSet.save(fn)